* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Many Body Quantum Mechanics
History of quantum field theory wikipedia , lookup
Perturbation theory (quantum mechanics) wikipedia , lookup
Quantum field theory wikipedia , lookup
Second quantization wikipedia , lookup
Quantum key distribution wikipedia , lookup
Identical particles wikipedia , lookup
Quantum teleportation wikipedia , lookup
Quantum decoherence wikipedia , lookup
Quantum entanglement wikipedia , lookup
EPR paradox wikipedia , lookup
Noether's theorem wikipedia , lookup
Measurement in quantum mechanics wikipedia , lookup
Probability amplitude wikipedia , lookup
Interpretations of quantum mechanics wikipedia , lookup
Quantum group wikipedia , lookup
Coherent states wikipedia , lookup
Molecular Hamiltonian wikipedia , lookup
Path integral formulation wikipedia , lookup
Bell's theorem wikipedia , lookup
Theoretical and experimental justification for the Schrödinger equation wikipedia , lookup
Relativistic quantum mechanics wikipedia , lookup
Hilbert space wikipedia , lookup
Hidden variable theory wikipedia , lookup
Quantum state wikipedia , lookup
Density matrix wikipedia , lookup
Canonical quantization wikipedia , lookup
Symmetry in quantum mechanics wikipedia , lookup
Self-adjoint operator wikipedia , lookup
Many Body Quantum Mechanics Jan Philip Solovej Notes Summer 2007 DRAFT of July 6, 2007 1 2 Contents 1 Preliminaries: Hilbert Spaces and Operators 1.1 4 Tensor products of Hilbert spaces . . . . . . . . . . . . . . . . . . 8 2 The Principles of Quantum Mechanics 2.1 11 Many body quantum mechanics . . . . . . . . . . . . . . . . . . . 14 3 Semi-bounded operators and quadratic forms 17 4 Extensions of operators and quadratic forms 20 5 Schrödinger operators 25 6 The canonical and grand canonical picture and the Fock spaces 32 7 Second quantization 35 8 One- and two-particle density matrices for bosonic or fermionic states 41 8.1 Two-particle density matrices . . . . . . . . . . . . . . . . . . . . 45 8.2 Generalized one-particle density matrix . . . . . . . . . . . . . . . 51 9 Bogolubov transformations 53 10 Quasi-free states 65 11 Quadratic Hamiltonians 68 12 Generalized Hartree-Fock Theory 71 13 Bogolubov Theory 73 13.1 The Bogolubov approximation . . . . . . . . . . . . . . . . . . . . A Extra Problems 75 100 A.1 Problems to Section 1 . . . . . . . . . . . . . . . . . . . . . . . . 100 A.2 Problems to Section 2 . . . . . . . . . . . . . . . . . . . . . . . . 101 A.3 Problems to Section 3 . . . . . . . . . . . . . . . . . . . . . . . . 102 3 A.6 Problems to Section 6 . . . . . . . . . . . . . . . . . . . . . . . . 103 B The Banach-Alaoglu Theorem 103 C Proof of the min-max principle 104 D Analysis of the function G(λ, Y ) in (37) 108 E Results on conjugate linear maps 109 F The necessity of the Shale-Stinespring condition 112 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 1 4 Preliminaries: Hilbert Spaces and Operators The basic mathematical objects in quantum mechanics are Hilbert spaces and operators defined on them. In order to fix notations we briefly review the definitions. 1.1 DEFINITION (Hilbert Space). A Hilbert Space H is a vector space endowed with a sesquilinear map (·, ·) : H × H → C (i.e., a map which is conjugate linear in the first variable and linear in the second1 ) such that kφk = (φ, φ)1/2 defines a norm on H which makes H into a complete metric space. 1.2 REMARK. We shall mainly use the following two properties of Hilbert spaces. (a) To any closed subspace V ⊂ H there corresponds the orthogonal complement V ⊥ such that V ⊕ V ⊥ = H. (b) Riesz representation Theorem: To any continuous linear functional Λ : H → C there is a unique ψ ∈ H such that Λ(φ) = (ψ, φ) for all φ ∈ H. We denote by H∗ the dual of the Hilbert space H, i.e., the space of all continuous linear functionals on H. The map J : H → H∗ defined by J(ψ)(φ) = (ψ, φ) is according to Riesz representation Theorem an anti-linear isomorphism. That J is anti-linear (or conjugate-linear) means that J(αφ + βψ) = αJ(φ) + βJ(ψ) for α, β ∈ C and φ, ψ ∈ H. We shall always assume that our Hilbert spaces are separable and therefore that they have countable orthonormal bases. We will assume that the reader is familiar with elementary notions of measure theory, in particular the fact that L2 - spaces are Hilbert spaces. 1.3 DEFINITION (Operators on Hilbert spaces). By an operator (or more precisely densely defined operator) A on a Hilbert space H we mean a linear map A : D(A) → H defined on a dense subspace D(A) ⊂ H. Dense refers to the fact that the norm closure D(A) = H. 1.4 DEFINITION (Extension of operator). If A and B are two operators such that D(A) ⊆ D(B) and Aφ = Bφ for all φ ∈ D(A) then we write A ⊂ B and say that B is an extension of A. 1 This is the convention in physics. In mathematics the opposite convention is used. JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 5 Note that the domain is part of the definition of the operator. In defining operators one often starts with a domain which turns out to be too small and which one then later extends. 1.5 DEFINITION (Symmetric operator). We say that A is a symmetric operator if (ψ, Aφ) = (Aψ, φ) (1) for all φ, ψ ∈ D(A). The result in the following problem is of great importance in quantum mechanics. 1.6 PROBLEM. Prove that (1) holds if and only if (ψ, Aψ) ∈ R for all ψ ∈ D(A). 1.7 REMARK. It is in general not easy to define the sum of two operators A and B. The problem is that the natural domain of A + B would be D(A) ∩ D(B), which is not necessarily densely defined. 1.8 EXAMPLE. The Hilbert space describing a one-dimensional particle without internal degrees of freedom is L2 (R), the space of square (Lebesgue) integrable functions defined modulo sets of measure zero. The inner product on L2 (R) is given by Z (g, f ) = g(x)f (x) dx. R The operator describing the kinetic energy is the second derivative operator. 2 d A = − 21 dx 2 defined originally on the subspace D(A) = C02 (R) = f ∈ C 2 (R) : f vanishes outside a compact set . Here C 2 (R) refers to the twice continuously differentiable functions on the real line. The subscript 0 refers to the compact support. The operator A is symmetric, since for φ, ψ ∈ D(A) we have by integration by parts Z 1 d2 φ ψ(x) 2 (x)dx (ψ, Aφ) = − 2 R dx Z Z 2 1 dψ dφ 1 dψ = (x) (x)dx = − (x)φ(x)dx = (Aψ, φ). 2 R dx dx 2 R dx2 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 6 1.9 DEFINITION (Bounded operators). An operator A is said to be bounded on the Hilbert Space H if D(A) = H and A is continuous, which by linearity is equivalent to kAk = sup kAφk < ∞. φ, kφk=1 The number kAk is called the norm of the operator A. An operator is said to be unbounded if it is not bounded. 1.10 PROBLEM. (a) Show that if an operator A with dense domain D(A) satisfies kAφk ≤ M kφk for all φ ∈ D(A) for some 0 ≤ M < ∞ then A can be uniquely extended to a bounded operator. (b) Show that the kinetic energy operator A from Example 1.8 cannot be extended to a bounded operator on L2 (R). 1.11 DEFINITION (Adjoint of an operator). If A is an operator we define the adjoint A∗ of A to be the linear map A∗ : D(A∗ ) → H defined on the space o n ∗ sup |(φ, Aψ)| < ∞ D(A ) = φ ∈ H ψ∈D(A), kψk=1 and with A∗ φ defined such that (A∗ φ, ψ) = (φ, Aψ) for all ψ ∈ D(A). The existence of A∗ φ for φ ∈ D(A∗ ) is ensured by the Riesz representation Theorem (why?). If D(A∗ ) is dense in H then A∗ is an operator on H. 1.12 PROBLEM. Show that the adjoint of a bounded operator is a bounded operator. 1.13 PROBLEM. Show that A is symmetric if and only if A∗ is an extension of A, i.e., A ⊂ A∗ . 1.14 EXAMPLE (Hydrogen atom). One of the most basic examples in quantum mechanics is the hydrogen atom. In this case the Hilbert space is H = L2 (R3 ; C2 ), i.e., the square integrable functions on R3 with values in C2 . Here C2 represents the internal spin degrees of freedom. The inner product is Z (g, f ) = g(x)∗ f (x) dx, R3 7 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 The total energy operator is2 H = − 12 ∆ − 1 , |x| (2) where ∆ = ∂12 + ∂22 + ∂32 , is the Laplacian. I.e., (Hφ)(x) = − 12 ∆φ(x) − 1 φ(x) |x| The domain of H may be chosen to be (3) D(H) = C02 R3 ; C2 = f ∈ C 2 (R3 ; C2 ) : f vanishes outside a compact subset of R3 . It is easy to see that if φ ∈ D(H) then Hφ ∈ H. It turns out that one may extend the domain of H to the Sobolev space H 2 (R3 ). We will return to this later. 1.15 EXAMPLE (Schrödinger operator). We may generalize the example of hydrogen to operators on L2 (Rn ) or L2 (Rn ; Cq ) of the form − 21 ∆ + V (x) where V : Rn → R is a potential. If V is a locally square integrable function we may start with the domain C02 (Rn ) or C02 (Rn ; C2 ). We shall return to appropriate conditions on V later. We call an operator of this form a Schrödinger operator. See Section 5 1.16 DEFINITION (Compact, trace class, and Hilbert-Schmidt operators). A linear operator K is said to be a compact operator on a Hilbert space H if D(K) = H and there are orthonormal bases u1 , u2 , . . . and v1 , v2 , . . . for H and a sequence λ1 , λ2 , . . . with limn→∞ λn = 0 such that Kφ = ∞ X λn (un , φ)vn (4) n=1 for all φ ∈ H. A compact operator K is said to be trace class if P 2 and it is called Hilbert-Schmidt if ∞ n=1 |λn | < ∞. P∞ n=1 |λn | < ∞ If K is trace class the trace of K is defined to be TrK = ∞ X λn (un , vn ). n=1 2 We use units in which Planck’s constant ~, the electron mass me , and the electron charge e are all equal to unity JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 8 1.17 PROBLEM. (a) Show that a trace class operator is Hilbert-Schmidt. (b) Show that the trace of a trace class operator on a Hilbert space H is finite and that if φ1 , φ2 , . . . is any orthonormal basis for H and K is any trace class operator on H then TrK = ∞ X (φn , Kφn ). n=1 1.18 PROBLEM. (a) (Super symmetry) Show that if K is a compact operator then K ∗ K and KK ∗ have the same non-zero eigenvalues with the same (finite) multiplicities. (b) Show that if K is a compact operator then it maps the eigenspaces of K ∗ K corresponding to non-zero eigenvalues to the eigenspace of KK ∗ with the same eigenvalue. (c) (Spectral Theorem for compact operators) Show that if K is a compact symmetric operator on a Hilbert space H then there is an orthonormal basis u1 , u2 , . . . for H and a sequence λ1 , λ2 , . . . ∈ R such that limn→∞ λn = 0 and Kφ = ∞ X λn (un , φ)un . n=1 (Hint: Diagonalize the finite dimensional operator obtained by restricting K to a non-zero eigenvalue eigenspace of K ∗ K = K 2 .) 1.1 Tensor products of Hilbert spaces Let H and K be two Hilbert spaces. The tensor product of H and K is a Hilbert space denoted H ⊗ K together with a bilinear map H × K 3 (u, v) 7→ u ⊗ v ∈ H ⊗ K, such that the inner products satisfy (u1 ⊗ v1 , u2 ⊗ v2 )H⊗K = (u1 , u2 )H (v1 , v2 )K , and such that the span{u ⊗ v | u ∈ H, v ∈ K} is dense in H ⊗ K. We call the vectors of the form u ⊗ v for pure tensor products. b is another tensor The tensor product is unique in the sense that if H⊗K b 7→ u⊗v extends uniquely to an isometric isomorphism. product then the map u⊗v 9 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 1.19 PROBLEM. Prove the above uniqueness statement. If (uα )α∈I is an orthonormal basis for H and (vβ )β∈J is an orthonormal basis for K, then (uα ⊗ vβ )α∈Iβ∈J is an orthonormal basis for H ⊗ K. 1.20 PROBLEM (Construction of the tensor product). Show that the tensor product H ⊗ K may be identified with the space `2 (I × J) and (u ⊗ v)ij = (ui , u)H (vj , v)K . More generally, if µ is a σ-finite measure on a measure space X and ν is a σ-finite measures on a measure space Y , it follows from Fubini’s Theorem that the tensor product L2 (X, µ) ⊗ L2 (Y, ν) may be identified with L2 (X × Y, µ × ν) (where µ × ν is the product measure) and f ⊗ g(x, y) = f (x)g(y). 1.21 PROBLEM. Use Fubini’s Theorem to show that L2 (X × Y, µ × ν) in this way may be identified with L2 (X, µ) ⊗ L2 (Y, ν). If we have an operator A on the Hilbert space H and an operator B on the Hilbert space K then we may form the tensor product operator A ⊗ B on H ⊗ K with domain D(A ⊗ B) = span {φ ⊗ ψ | φ ∈ D(A), ψ ∈ D(B)} and acting on pure tensor products as A ⊗ B(φ ⊗ ψ) = (Aφ) ⊗ (Bψ). The tensor product may in a natural way be extended to more than two Hilbert spaces. In particular, we may for N = 1, 2, . . . consider the N -fold tensor N product N H of a Hilbert space H with itself. On this space we have a natural action of the symmetric group SN . I.e., if σ ∈ SN then we have a unitary map N N Uσ : N H → N H defined uniquely by the following action on the pure tensor products Uσ u1 ⊗ · · · ⊗ uN = uσ(1) ⊗ · · · ⊗ uσ(N ) . We shall denote by Ex : H ⊗ H → H ⊗ H the unitary corresponding to a simple interchange of the two tensor factors. 10 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 1.22 PROBLEM. Show that Uσ defines a unitary operator and that the two operators P+ = (N !)−1 X Uσ , P− = (N !)−1 X (−1)σ Uσ , (5) σ∈SN σ∈SN ∗ 2 are orthogonal projections (P = P , P = P ) satisfying P− P+ = 0 if N ≥ 2. Here (−1)σ is the sign of the permutation σ. Correction since April 15: N ≥ 2 The two projections P± define two important subspaces of NN added H. Correction since May 3: definition 1.23 DEFINITION (Symmetric and anti-symmetric tensor products). The corrected H := P+ N O H . (6) H . (7) sym The antisymmetric tensor product is the space N ^ H := P− correction 6: symmetric tensor product is the space N O of (−1)σ added N O We define the antisymmetric tensor product of the vectors u1 , . . . , uN ∈ H as u1 ∧ · · · ∧ uN = (N !)1/2 P− (u1 ⊗ · · · ⊗ uN ). (8) 1.24 PROBLEM. Show that if u1 , . . . , uN are orthonormal then u1 ∧ · · · ∧ uN is normalized (i.e., has norm 1). 1.25 PROBLEM. Let u1 , . . . , uN be orthonormal functions in an L2 space L2 (X, µ) over the measure space X with measure µ. Show that in the space L2 (X N , µN ) u1 (x1 ) · · · uN (x1 ) u1 (x2 ) · · · uN (x2 ) u1 ∧ · · · ∧ uN (x1 , . . . , xN ) = (N !)−1/2 det .. . u1 (xN ) · · · uN (xN ) One refers to this as a Slater determinant. 1.26 PROBLEM. Show that if dim H < N then VN H = {0}. . July sentence JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 11 1.27 PROBLEM. Let X be a measure space with measure µ and let H be a Hilbert space. Show that we may identify the tensor product L2 (X, µ) ⊗ H with the Hilbert space L2 (X, µ; H) of H-valued L2 functions on X, where the tensor product of f ∈ L2 (X, µ) with u ∈ H is the function f ⊗ u(x) = f (x)u. 2 The Principles of Quantum Mechanics We shall here briefly review the principles of quantum mechanics. The reader with little or no experience in quantum mechanics is advised to also consult standard textbooks in physics. In quantum mechanics a pure state of a physical system is described by a unit vector ψ0 in a Hilbert space H. The measurable quantities correspond to ‘expectation values’ hAiψ0 = (ψ0 , Aψ0 ), of operators A on H. Of course, in order for this to make sense we must have ψ0 ∈ D(A). Since measurable quantities are real the relevant operators should have real expectation values, i.e, the operators are symmetric. (See Problem 1.6). The physical interpretation of the quantity hAiψ0 is that it is the average value of ‘many’ measurements of the observable described by the operator A in the state ψ0 . As an example ψ0 ∈ C02 (R3 ; C2 ) with R |ψ0 |2 = 1 may represent a state of a hydrogen atom (see Example 1.14). The average value of many measurements of the energy of the atom in this state will be Z 1 1 1 ψ0 , (− 2 ∆ − )ψ0 = ψ0 (x)∗ (− 21 ∆ − )ψ0 (x) dx |x| |x| 3 ZR 1 2 1 |∇ψ (x)| − = |ψ0 (x)|2 dx, 0 2 |x| R3 where the last equality follows by integration by parts. The general quantum mechanical state, which is not necessarily pure is a statistical average of pure states, i.e, expectations are of the form hAi = ∞ X n=1 λn (ψn , Aψn ), (9) JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 where 0 ≤ λn ≤ 1 with P n 12 λn = 1 and ψn is a family of orthonormal vectors. In this representation the λn are unique3 . How far the state is from being pure is Correction April naturally measured by its entropy. → 15: since unit orthonormal. Uniqueness statement. 2.1 DEFINITION (Von Neumann entropy). The von Neumann entropy of a Problem added to footnote state h·i of the form (9) is S(h·i) = − ∞ X λn log λn , n=1 which is possibly +∞. (We use the convention that t log t = 0 if t = 0.) Addition since April 15: which is possibly +∞ Note that the entropy vanishes if and only if the state is pure. Of particular interest are the equilibrium states, either zero (absolute) temperature or positive temperature states. The zero temperature state is usually a pure state, i.e., given by one vector, whereas the positive temperature states (the Gibbs states) are non-pure. Both the zero temperature states and the positive temperature states are described in terms of the energy operator, the Hamiltonian. We shall here mainly deal with the zero temperature equilibrium states, the ground states. 2.2 DEFINITION (Stability and Ground States). Consider a physical system described by a Hamiltonian, i.e., energy operator, H on a Hilbert space H. If inf φ∈D(H), kφk=1 (φ, Hφ) > −∞ the system is said to be stable. If this holds we call E= inf φ∈D(H), kφk=1 (φ, Hφ) for the ground state energy. A ground state for the system, if it exists, is a unit vector ψ0 ∈ D(H) such that (ψ0 , Hψ0 ) = inf φ∈D(H), kφk=1 (φ, Hφ) . Thus a ground state is characterized by minimizing the energy expectation. JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 13 2.3 DEFINITION (Free energy and temperature states). The free energy of a Correction since April 15: stable added stable system at temperature T ≥ 0 is F (T ) = inf (hHi − T S(h·i)) , h·i hHi<∞ (possibly −∞) where the infimum is over all states of the form (9) with ψn ∈ D(H) for n = 1, 2, . . .. If for T > 0 a minimizer exists for the free energy variation above it is called a Gibbs state at temperature T . 2.4 PROBLEM. Show that F (T ) is a decreasing function of T and that F (0) = E, i.e., the free energy at zero temperature is the ground state energy. 2.5 PROBLEM (Ground state eigenvector). Show that if ψ0 is a ground state with (ψ0 , Hψ0 ) = λ then Hψ0 = λψ0 , i.e., ψ0 is an eigenvector of H with eigenvalue λ. (Hint: consider the normalized vector φε = ψ0 + εφ kψ0 + εφk for φ ∈ D(H). Use that the derivative of (φε , Hφε ) wrt. ε is zero at ε = 0.) 2.6 PROBLEM (Stability of free particle). Show that the free 1-dimensional particle described in Example 1.8 is stable, but does not have a ground state. Show that its free energy is F (T ) = −∞ for all T > 0. 2.7 PROBLEM (Gibbs state). Show that if h·i is a Gibbs state at temperature T > 0 then Tr(A exp(−H/T )) Tr(exp(−H/T )) for all bounded operators A. In particular, exp(−H/T ) is a trace class operator. hAi = (Hint: Use Jensen’s inequality and the fact that t 7→ t log t is strictly convex. The problem is easier if one assumes that exp(−H/T ) is trace class, otherwise some version of the spectral Theorem is needed4 . ) Correction since April 15: footnote The Hamiltonian H for hydrogen, given in (2) and (3), is stable, it does not have a ground state on the domain C02 , but in this case, however, this is simply added Correction since April 15: ∞ → 2 3 More generally, a state may be defined as a normalized positive linear functional on the bounded operators on H (or even on some other algebra of operators). Here we shall only consider states of the form (9) (see also Problem A.2.2). 4 Theorem 4.12 is sufficient 14 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 because the domain is too small (see Section 5). On the extended domain H 2 (R3 ) the Hamiltonian does have a ground state. Finding the correct domain on which a Hamiltonian has a possible ground state is an important issue in quantum mechanics. In Section 3 we discuss in some generality operators and quadratic forms. We shall only be concerned with the eigenvalues of the operators and not with the continuous part of the spectrum. We therefore do not need to understand the Spectral Theorem in its full generality and we shall not discuss it here. We therefore do not need to understand the more complex questions concerning selfadjointness. We mainly consider semi bounded operators and the corresponding quadratic forms. The notion of quadratic forms is very essential in quantum mechanics. As we have seen the measurable quantities corresponding to an observable, represented by an operator A are the expectation values which are of the form (ψ, Aψ). In applications to quantum mechanics it is therefore relevant to try to build the general theory as much as possible on knowledge of these expectation values. The map ψ 7→ (ψ, Aψ) is a special case of a quadratic form. 2.1 Many body quantum mechanics Consider N quantum mechanical particles described on Hilbert spaces h1 , . . . , hN and with Hamilton operators h1 , . . . , hN . The combined system of these particles is described on the tensor product HN = h1 ⊗ · · · ⊗ hN . We may identify the operators h1 , . . . , hN with operators on this tensor product space. I.e., we identify h1 , . . . , hN with the operators h1 ⊗ I ⊗ · · · ⊗ I, I ⊗ h2 ⊗ I ⊗ · · · ⊗ I, ... I ⊗ I ⊗ · · · ⊗ hN . If the particles are non-interacting the Hamiltonian operator for the combined system is simply HNin = h1 + . . . + hN . This operator may be defined on the domain D(HNin ) = span{φ1 ⊗ · · · ⊗ φN | φ1 ∈ D(h1 ), . . . , φN ∈ D(hN )}. 15 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Correction 2.8 PROBLEM. Show that if D(h1 ), . . . , D(hN ) are dense in h1 , . . . , hN respec- since April 15: one “if” removed 1 → N tively then D(HNin ) is dense in HN . 2.9 THEOREM (ground state of non-interacting particles). If ej = inf φ∈D(hj ),kφk=1 (φ, hj φ), j = 1, . . . , N are ground state energies of the Hamiltonians h1 , . . . , hN then the ground state P energy of HNin is N j=1 ej . Moreover, if φ1 , . . . , φN are ground state eigenvectors of h1 , . . . , hN then φ1 ⊗ · · · ⊗ φN is a ground state eigenvector for HN . Correction since April 15: H → H Proof. If Ψ ∈ D(HNin ) is a unit vector we may write Ψ = ψ1 ⊗ Ψ1 + . . . + ψK ⊗ ΨK where ψ1 , . . . , ψK ∈ D(h1 ) and Ψ1 , . . . , ΨK ∈ h2 ⊗ · · · ⊗ hN are orthonormal. Since Ψ is a unit vector we have kψ1 k2 + . . . + kψK k2 = 1. We have (Ψ, h1 Ψ) = K X (ψi , h1 ψi ) ≥ i=1 Hence (Ψ, HNin Ψ) ≥ K X kψi k2 e1 = e1 . i=1 PN j=1 ej . On the other hand if we, given ε > 0, choose unit vectors φj ∈ D(hj ), j = 1, . . . , N such that (φj , hj φj ) < ej +ε for j = 1, . . . , N and define Ψ = φ1 ⊗· · ·⊗φN . We find that Ψ is a unit vector and (Ψ, HNin Ψ) = N X (φj , hj φj ) ≤ j=1 N X ej + N ε. j=1 It is clear that if φ1 , . . . , φN are ground state eigenvectors for h1 , . . . , hN then Ψ is a ground state eigenvector for HNin . The physically more interesting situation is for interacting particles. The most common type of interactions is for two particles to interact pairwise. We talk about 2-body interactions. The interaction of particle i and particle j (i < j say) is described by an operator Wij acting in the Hilbert space hi ⊗ hj . As we shall now explain we may again identify such an operator with an operator on Correction since May 3: N → K 16 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 h1 ⊗ · · · ⊗ hN , which we also denote by Wij . Let us for simplicity of notation assume that i = 1 and j = 2. We then identify W12 with the operator W12 ⊗ I ⊗ · · · ⊗ I (the number of identity operators in this tensor product is N − 2) thinking of h1 ⊗ · · · ⊗ hN = (h1 ⊗ h2 ) ⊗ · · · ⊗ hN . The interacting Hamiltonian is then formally HN = HNin + X Wij = 1≤i<j≤N N X X hj + j=1 Wij . 1≤i<j≤N The reason this is only formal is that the domain of the operator has to be specified and it may depend on the specific situation. Determining the ground state energy and possible ground state eigenfunctions of an interacting many particle quantum Hamiltonian is a very difficult problem. It can usually not be done exactly and different approximative methods have been developed and we shall discuss these later. Finally, we must discuss one of the most important issues of many body quantum mechanics. The question of statistics of identical particles. Assume that the N particles discussed above are identical, i.e., h1 = . . . = hN = h, h1 = . . . = hN = h. If the particles are interacting we also have that the 2-body potential Wij is the same operator W for all i and j and that ExW Ex = W , where Ex is the unitary exchange operator. When we identify the 1-body Hamiltonian h and the 2-body potential W with operators on h ⊗ · · · ⊗ h we must still write subscripts on them: hj and Wij . This is to indicate on which of the tensor factors they act, e.g. h1 = h ⊗ I ⊗ · · · ⊗ I, W12 = W ⊗ I ⊗ · · · ⊗ I. It is now clear that the non-interacting operator HNin maps vectors in the subN VN spaces N h into the same subspaces. The operator may therefore sym h and be restricted to the domains P+ D(HNin ) or P− D(HNin ). Correction since May “by” 3: inserted 17 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 If we restrict to the symmetric subspace NN sym h we refer to the particles as bosons and say that they obey Bose-Einstein statistics. If we restrict to the V antisymmetric subspace N h we refer to the particles as fermions and say that they obey Fermi-Dirac statistics. As we shall see the physics is very different for these two types of systems. The interaction Hamiltonian will also formally map the subspaces VN NN sym h and h into themselves. This is only formal since we have not specified the domain of the interaction Hamiltonian. We have an immediate corollary to Theorem 2.9. 2.10 COROLLARY (Ground state of Bose system). We consider the HamilN tonian HNin for N identical particles restricted to the symmetric subspace N sym h, i.e., with domain Dsym (HNin ) = P+ D(HNin ). The ground state energy of this bosonic system is N e if e is the ground state energy of h. Moreover, if h has a ground state eigenvector φ then HNin has the ground state eigenvector φ ⊗ · · · ⊗ φ. Note in particular that the ground state energy of HNin on the symmetric NN N h. subspace N sym h is the same as on the full Hilbert space The situation for fermions is more complicated and we will return to it later. 2.11 EXAMPLE (Atomic Hamiltonian). The Hamilton operator for N electrons in an atom with nuclear charge Z and with the nucleus situated at the origin is N X i=1 (− 21 ∆i − X Z 1 )+ . |xi | |x i − xj | 1≤i<j≤N Since physical electrons are fermions this Hamiltonian should be considered on V the antisymmetric Hilbert space N L2 (R3 ; C2 ). We shall return to showing that an atom is stable. 3 Semi-bounded operators and quadratic forms 3.1 DEFINITION (Positive Operators). An operator defined on a subspace D(A) of H is said to be positive (or positive definite) if (ψ, Aψ) > 0 for all JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 18 non-zero ψ ∈ D(A). It is said to be positive semi-definite if (ψ, Aψ) ≥ 0 for all ψ ∈ D(A). In particular, such operators are symmetric. The notion of positivity induces a partial ordering among operators. 3.2 DEFINITION (Operator ordering). If A and B are two operators with D(A) = D(B)5 then we say that A is (strictly) less than B and write A < B if the operator B − A (which is defined on D(B − A) = D(A) = D(B) is a positive definite operator. We write A ≤ B if B − A is positive semi-definite. 3.3 DEFINITION (Semi bounded operators). An operator A is said to be bounded below if A ≥ −cI for some scalar c. Here I denotes the identity operator on H. Likewise an operator A is said to be bounded above if A ≤ cI. 3.4 DEFINITION (Quadratic forms). A quadratic form Q is a mapping Q : D(Q)×D(Q) → C (where D(Q) is a (dense) subspace of H), which is sesquilinear (conjugate linear in the first variable and linear in the second): Q(α1 φ1 + α2 φ2 , ψ) = α1 Q(φ1 , ψ) + α2 Q(φ2 , ψ) Q(φ, α1 ψ1 + α2 ψ2 ) = α1 Q(φ, ψ1 ) + α2 Q(φ, ψ2 ). We shall often make a slight abuse of notation and denote Q(φ, φ) by Q(φ). A quadratic form Q is said to be positive definite if Q(φ) > 0 for all φ 6= 0 and positive semi-definite if Q(φ) ≥ 0. It is said to be bounded below if Q(φ) ≥ −ckφk2 (and above if Q(φ) ≤ ckφk2 ) for some c ∈ R. 3.5 PROBLEM (Cauchy-Schwarz inequality). Show that if Q is a positive semidefinite quadratic form it satisfies the Cauchy-Schwarz inequality |Q(φ, ψ)| ≤ Q(φ)1/2 Q(ψ)1/2 . (10) 3.6 DEFINITION (Bounded quadratic forms). A quadratic form Q is said to be bounded if there exists 0 ≤ M < ∞ such that |Q(φ)| ≤ M kφk2 , for all φ ∈ D(Q). 5 It would be more correct to say that we require that D(A) ∩ D(B) is dense and that B − A is positive (semi)-definite on this subspace Correction since April 15: footnote added 19 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Note that quadratic forms that are bounded above and below are bounded, but the converse is not true since bounded quadratic forms are not necessarily real. As for operators we have that if Q is positive semi-definite (or even just bounded above or below) then it is symmetric, meaning Q(φ, ψ) = Q(ψ, φ). (11) The proof is the same as in Problem 1.6. 3.7 PROBLEM. Show that if Q is a bounded quadratic form then it extends to a unique bounded quadratic form on all of H (compare Problem 1.10). 3.8 PROBLEM. Show that if Q is a quadratic form then it is enough to know, Q(φ) = Q(φ, φ) for all φ ∈ D(Q), in order to determine Q(ψ1 , ψ2 ) for all ψ1 , ψ2 ∈ D(Q). It is clear that to an operator A we have a corresponding quadratic form Correction since April 15: “All” added Q(φ) = (φ, Aφ). The next problem shows that the opposite is also true. 3.9 PROBLEM (Operators corresponding to quadratic forms). Show that corresponding to a quadratic form there exists a unique linear map A : D(A) → H, with ( D(A) = ) |Q(ψ, φ)| φ ∈ D(Q) : sup <∞ kψk ψ∈D(Q)\{0} Correction since April 15: operator→ linear map. Formula- tion improved such that Q(ψ, φ) = (ψ, Aφ) for all φ ∈ D(A) and ψ ∈ D(Q). Note, that we may to have that D(A) is a strict subspace of D(Q). In fact, in general D(A) need not φ Q(ψ, φ) (ψ, Aφ) ∈ for = all D(A) and ψ ∈ D(Q) even be dense (see Example 5.4). The quadratic form corresponding to a Schrödinger operator − 21 ∆ + V is Z Z 1 Q(φ) = − 2 φ(x)∆φ(x)dx + V (x)|φ(x)|2 dx Z Z 2 1 = 2 |∇φ(x)| dx + V (x)|φ(x)|2 dx for φ being a C02 function. In Section 11.3 in Lieb and Loss Analysis6 conditions Correction since are given on the potential V that ensure that the quadratic form corresponding to a Schrödinger operator can be extended to the Sobolev space H 1 (R3 ). 6 Lieb and Loss Analysis, AMS Graduate Studies in Mathematics Vol. 114 2nd edition April C0∞ → C02 15: JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 20 3.10 PROBLEM. Assume that Q is quadratic form, which is bounded below and that A is the corresponding operator defined in Problem 3.9. If a unit vector ψ0 ∈ D(Q) satisfies that Q(ψ0 ) = inf Q(φ) φ∈D(Q),kφk=1 show that ψ0 is a ground state eigenvector for A. Theorem 11.5 in Lieb and Loss Analysis gives conditions ensuring that a Schrödinger operator has a ground state. 4 Extensions of operators and quadratic forms We shall here briefly sketch how to define a natural extension of a symmetric operator and how to define a natural extension of the corresponding quadratic form if the operator is bounded below. 4.1 DEFINITION (Closed operator). An operator A on a Hilbert space H is said to be closed if its graph Correction since April 15: ∈ H ⊕H G(A) = {(φ, Aφ) ∈ H ⊕ H | φ ∈ D(A)} is closed in the Hilbert space H ⊕ H. 4.2 THEOREM (Closability of symmetric operator). If A is a symmetric (densely defined) operator on a Hilbert space H then the closure of its graph G(A) is the graph of a closed operator A, the closure of A. Proof. We have to show that we can define an operator A with domain D(A) = {φ ∈ H | ∃ψ ∈ H : (φ, ψ) ∈ G(A)} such that for φ ∈ D(A) we have Aφ = ψ if (φ, ψ) ∈ G(A). The only difficulty in proving that this defines a closed (linear) operator is to show that there is at most one ψ for which (φ, ψ) ∈ G(A). Thus we have to show that if (0, ψ) ∈ G(A) then ψ = 0. If (0, ψ) ∈ G(A) we have a sequence φn ∈ D(A) with limn→∞ φn = 0 and limn→∞ Aφn = ψ. For all φ0 ∈ D(A) we then have since A is symmetric that (φ0 , ψ) = lim (φ0 , Aφn ) = lim (Aφ0 , φn ) = 0. n→∞ n→∞ added JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 21 Thus ψ ∈ D(A)⊥ , but since D(A) is dense we have ψ = 0. 4.3 EXAMPLE. If we consider the Laplace operator with domain C02 (Rn ), then the domain of the closure is the Sobolev space H 2 (Rn ). Correction since 4.4 DEFINITION (Closed quadratic form). A quadratic form Q satisfying the April 15: C0∞ → C02 lower bound Q(φ) ≥ −αkφk2 for some α > 0 is said to be closed if the domain D(Q) is complete under the norm kφkα = p (α + 1)kφk2 + Q(φ). (That this is a norm follows from the Cauchy-Schwarz inequality in Problem 3.5.) 4.5 PROBLEM. Show that the above definition does not depend on how large α is chosen. Correction since 4.6 THEOREM (Closability of form coming from semibounded operator). Let April symmetric 15: → semibounded. A be a (densely defined) operator on a Hilbert space H with the lower bound Q(φ) ≥ −αkφk2 added. A ≥ −αI for some α > 0. Then there exists a unique closed quadratic form Q such that • D(A) ⊆ D(Q). • D(Q) is the closure of D(A) under the norm k · kα . • Q(φ) ≥ −αkφk2 • Q(φ) = (φ, Aφ) for φ ∈ D(A). Proof. We consider the norm kφkα = p (α + 1)kφk2 + (φ, Aφ) defined for φ ∈ D(A). Observe that kφk ≤ kφkα . Thus if φn ∈ D(A) is a Cauchy sequence for the norm k · kα it is also a Cauchy sequence for the original norm k · k. Hence there is a φ ∈ H such that limn→∞ φn = φ. Moreover, since k · kα is a norm it follows that kφn kα is a Cauchy sequence of real numbers, which hence converges to a real number. Since limn→∞ kφn k = kφk we conclude that the sequence (φn , Aφn ) converges to a real number. JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 22 We want to define the quadratic form Q having domain D(Q) consisting of all vectors φ ∈ H for which there is a Cauchy sequence φn ∈ D(A) under the k · kα norm such that limn→∞ φn = φ. For such a φ we define Q(φ) = limn→∞ (φn , Aφn ). The only difficulty in proving the theorem is to show that φ = 0 implies that limn→∞ (φn , Aφn ) = 0. In fact, all we have to show is that limn→∞ kφn kα = 0. Let us denote by (ψ 0 , ψ)α = (α + 1)(ψ 0 , ψ) + (ψ 0 , Aψ) the inner product corresponding to the norm k · kα . Then kφn k2α = (φn , φm )α + (φn , φn − φm )α ≤ |(φn , φm )α | + kφn kα kφn − φm kα . The second term tends to zero as n tends to infinity with m ≥ n since φn is a Cauchy sequence for the k · kα norm and kφn kα is bounded. For the first term Correction since April 15: absolute values inserted above we have since A is symmetric (φn , φm )α = (α + 1)(φn , φm ) + (φn , Aφm ) = (α + 1)(φn , φm ) + (Aφn , φm ). This tends to 0 as m tends to infinity since limm→∞ φm = φ = 0. 4.7 EXAMPLE. If we consider the Laplace operator with domain C02 (Rn ), then the domain of the closed quadratic form in the theorem above is the Sobolev space H 1 (Rn ). Correction since 4.8 DEFINITION (Friedrichs’ extension). The symmetric operator which ac- April 15: C0∞ → C02 cording to Problem 3.9 corresponds to the closed quadratic form Q described in Theorem 4.6 is called the Friedrichs’ extension of the operator A, we will denote it AF . Correction since We will in the future often prove results on conveniently chosen domains. These results may then by continuity be extended to the naturally extended domain for the Friedrichs’ extension. In particular we see that stable Hamiltonians H have a Friedrichs’ extension. 4.9 PROBLEM. Show that the Friedrichs’ extension of an operator is a closed operator and hence that A ⊆ AF . 4.10 PROBLEM. Argue that Friedrichs extending an operator that is already a Friedrichs extension does not change the operator, i.e., (AF )F = AF . April Notation added 15: AF JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 23 Hence the Friedrichs extension is in general a larger extension than the closure of the operator. In Problem 5.19 we shall see that the Friedrichs extension may in fact be strictly larger than the closure. 4.11 PROBLEM. Show that if A is bounded below then the Friedrichs’ extension of A + λI (defined on D(A)) for some λ ∈ R is (A + λI)F = AF + λI defined on D(AF ). We have seen that symmetric operators are characterized by A ⊆ A∗ . The Friedrichs’ extensions belong to the more restrictive class of self-adjoint operators satisfying A = A∗ . Self-adjoint operators are very important. It is for this class of operators that one has a general spectral theorem. We will here not discuss selfadjoint operators in general, but restrict attention to Friedrichs’ extensions. The closure of a symmetric operator is in general not self-adjoint. If it is the operator is called essentially self-adjoint. Short of giving the full spectral theorem we will in the next theorem characterize the part of the spectrum of a Friedrichs’ extension which corresponds to eigenvalues below the essential spectrum. We will not here discuss the essential spectrum, it includes the continuous spectrum but also eigenvalues of infinite multiplicity. In these lecture notes we will only be interested in aspects of physical systems which may be understood solely from the eigenvalues below the essential spectrum. For hydrogen the spectrum is the set 1 − 2 : n = 1, 2, . . . ∪ [0, ∞). 2n The essential spectrum [0, ∞) corresponds here to the continuous spectrum. The eigenvalues can be characterize as in the theorem below. In the next section we will do this for the ground state energy. 4.12 THEOREM (Min-max principle for Friedrichs’ extension). Consider an operator A which is bounded from below on a Hilbert space H. Define the sequence µn = µn (A) = inf max (φ, Aφ) : M ⊆ D(A), dimM = n . (12) φ∈M, kφk=1 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 24 Then µn is a non-decreasing sequence and unless µ1 , . . . , µk are eigenvalues of the Friedrichs’ extension AF of A counted with multiplicities we have µk = µk+1 = µk=2 = . . . . If this holds we call µk the bottom of the essential spectrum. Correction since If µk < µk+1 then the infimum above for n = k is attained for the k- May Formulation changed slightly dimensional space Mk spanned by the eigenfunctions of AF corresponding to the eigenvalues µ1 , . . . , µk in the sense that µk = max φ∈Mk , kφk=1 (φ, AF φ), If φ ∈ Mk⊥ ∩ D(AF ) then (φ, AF φ) ≥ µk+1 kφk2 . On the other hand if µ1 , . . . , µk are eigenvalues for AF with corresponding eigenvectors spanning a k-dimensional space Mk such that (φ, AF φ) ≥ µk kφk2 for all φ ∈ Mk⊥ ∩ D(Q) then (12) holds for n = 1, . . . , k. The proof is given in Appendix C. Note in particular that since A is assumed to be bounded from below µ1 (A) = inf φ∈D(A), kφk=1 (φ, Aφ) > −∞. (13) 4.13 PROBLEM. Show that if µn are the min-max values defined in Theorem 4.12 for an operator A which is bounded below then N X µn (A) = n=1 inf{Tr(P A) | P an orth. proj. onto an N-dimensional subspace of D(A)}. 4.14 PROBLEM (Operators with compact resolvent). Assume that the minmax values µn (A) of an operator A on a Hilbert space H which is bounded below satisfy µn (A) → ∞ as n → ∞. Show that we may choose an orthonormal basis of H consisting of eigenvectors of AF . Show that there is a constant α > 0 such that the Friedrichs’ extension of A + αI (defined on D(A)) is an injective operator that maps onto all of the Hilbert space. Show that the inverse map (A + αI)−1 is compact. The operator (A + αI)−1 is called a resolvent of A and we say that A has compact resolvent. 3: JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 5 25 Schrödinger operators We shall in this section discuss Schrödinger operators (see Example 1.15) in more details. 5.1 DEFINITION (Schrödinger operator on C02 (Rn )). The Schrödinger operator for a particle without internal degrees of freedom moving in a potential V ∈ L2loc (Rn )7 H = − 21 ∆ − V with domain D(H) = C02 (Rn ). As we saw earlier we also have the Schrödinger quadratic form. 5.2 DEFINITION (Schrödinger quadratic form on C01 (Rn )). The Schrödinger quadratic form for a particle without internal degrees of freedom moving in a potential V ∈ L1loc (Rn ) is Q(φ) = 1 2 Z 2 Z |∇φ| − Rn V |φ|2 Rn with domain D(Q) = C01 (Rn ). Note that in order to define the quadratic form on C01 (Rn ) we need only assume that V ∈ L1loc whereas for the operator we need V ∈ L2loc . 5.3 PROBLEM. If V ∈ L2loc show that the operator defined as explained in Problem 3.9 from the Schrödinger quadratic form Q with D(Q) = C01 (Rn ) is indeed an extension of the Schrödinger operator H = −∆ − V to a domain which includes C02 (Rn ). If V ∈ L1loc \ L2loc then this need not be the case as explained in the next example. 5.4 EXAMPLE. Consider the function f : Rn → R given by ( |x|−n/2 , if |x| < 1 f (x) = 0, otherwise Then f is in L1 (Rn ) but not in L2 (Rn ). Let q1 , q2 , . . . be an enumeration of the P rational points in Rn and define V (x) = i i−2 f (x − qi ). Then V ∈ L1loc (Rn ) but The space Lploc (Rn ), for some p ≥ 1 consists of functions f (defined modulo sets of measure R zero), such that C |f |p < ∞ for any compact set C ⊂ Rn 7 26 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 for all ψ ∈ C01 (Rn ) we have V ψ 6∈ L2 . This follows easily since |V (x)|2 |ψ(x)|2 ≥ i−2 |x−qi |−n |ψ(x)|2 for all i. Therefore the domain of the operator A defined from the Schrödinger quadratic form Q with D(Q) = C01 (Rn ) is D(A) = {0}. We shall now discuss ways of proving that the Schrödinger quadratic form is bounded from below. We begin with the Perron-Frobenius Theorem for the Schrödinger operator. Namely, the fact that if we have found a non-negative eigenfunction for the Schrödinger operator then the corresponding eigenvalue is the lowest possible expectation for the Schrödinger quadratic form. 5.5 THEOREM (Perron-Frobenius for Schrödinger). Let V ∈ L1loc (Rn ). Assume that 0 < ψ ∈ C 2 (Rn ) ∩ L2 (Rn ) and that (− 21 ∆ − V )ψ(x) = λψ(x) for all x in some open set Ω. Then for all φ ∈ C01 (Rn ) with support in Ω we have Z Z Z 2 2 1 Q(φ) = 2 |∇φ| − V |φ| ≥ λ |φ|2 . Rn Rn Rn Proof. Given φ ∈ C01 (Rn ) we can write φ = f ψ, where f ∈ C01 (Rn ). Then Z Z 2 2 2 2 1 V |f ψ|2 Q(φ) = 2 ψ |∇f | + |f | |∇ψ| + f ∇f + f ∇f ψ∇ψ − n n R Z ZR 2 2 2 1 V |f ψ| ≥ 2 |f | |∇ψ| + f ∇f + f ∇f ψ∇ψ − n Rn Z R Z 2 = |f | ψ(− 12 ∆ − V )ψ = λ |φ|2 , Rn Rn where the second to last identity follows by integration by parts. 5.6 COROLLARY (Lower bound on hydrogen). For all φ ∈ C01 (R3 ) we have Z Z Z Z2 2 −1 2 1 |∇φ(x)| dx − Z|x| |φ(x)| dx ≥ − |φ(x)|2 dx. 2 2 Proof. Consider the function ψ(x) = e−Z|x| . Then for all x 6= 0 we have (− 21 ∆ − Z|x|−1 )ψ(x) = − Z2 ψ(x). 2 The statement therefore immediately follows for all φ ∈ C01 (R3 ) with support away from 0 from the previous theorem. The corollary follows for all φ ∈ C01 (R3 ) using the result of the next problem. 27 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 5.7 PROBLEM. Show that all φ ∈ C01 (R3 ) can be approximated by functions φn ∈ C01 (R3 ) with support away from 0 in such a way that Z Z Z Z 2 −1 2 2 |∇φn (x)| dx − Z|x| |φn (x)| dx → |∇φ(x)| dx − Z|x|−1 |φ(x)|2 dx. 5.8 PROBLEM. Show that the function ψ(x) = e−Z|x| as a function in L2 (R3 ) is an eigenfunction with eigenvalue −Z 2 /2 for the Friedrichs’ extension of H = − 21 ∆ − Z|x|−1 (originally) defined on C02 (R3 ). It is rarely possible to find positive eigenfunctions. A much more general approach to proving lower bounds on Schrödinger quadratic forms is to use the Sobolev inequality. In a certain sense this inequality is an expression of the celebrated uncertainty principle. 5.9 THEOREM (Sobolev Inequality). For all φ ∈ C01 (Rn ) with n ≥ 3 we have the Sobolev inequality kφk 2n n−2 ≤ 2(n − 1) k∇φk2 n−2 Proof. Let u ∈ C01 (Rn ) then we have Z xi ∂i u(x1 , . . . , xi−1 , x0i , xi+1 , . . . , xn )dx0i . u(x) = −∞ Hence |u(x)| n n−1 ≤ n Z Y i=1 1 ! n−1 ∞ |∂i u|dxi . −∞ Thus by the general Hölder inequality (in the case n = 3 simply by CauchySchwarz) Z ∞ |u(x)| n n−1 Z dx1 ≤ −∞ 1 n−1 ∞ n Z Y |∂1 u|dx1 −∞ i=2 ∞ Z −∞ 1 ! n−1 ∞ |∂i u|dx1 dxi −∞ Using the same argument for repeated integrations over x2 , . . . , xn gives Z |u(x)| Rn n n−1 dx ≤ n Z Y i=1 1 ! n−1 |∂i u|dx Rn Thus n kuk n−1 ≤ k∇uk1 . . . 28 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Now set u = φ 2(n−1) n−2 . (The reader may at this point worry about the fact that u is not necessarily C 1 . One can easily convince oneself that the above argument works for this u too. Alternatively , in the case n = 3 which is the one of interest here 2(n−1) n−2 is an integer and thus u is actually C 1 .) We then get kφk 2(n−1) n−2 2n n−2 ≤ n 2(n − 1) kφk n−2 2n k∇φk2 . n−2 n−2 Especially for n = 3 we get kφk6 ≤ 4k∇φk2 . The sharp constant in the Sobolev inequality was found by Talenti8 In the case n = 3 the sharp version of the Sobolev inequality is √ 3 (2π 2 )1/3 k∇φk2 ≈ 2.34k∇φk2 . kφk6 ≤ 2 (14) 5.10 THEOREM (Sobolev lower bound on Schrödinger). Assume that V ∈ L1loc (Rn ), n ≥ 3 and that the positive part V+ = max{V, 0} of the potential satisfies V+ ∈ L 2+n 2 (Rn ). Then for all φ ∈ C01 (Rn ) we get Z Z 2 1 Q(φ) = 2 |∇φ| − V |φ|2 Rn 2 ≥ − n+2 Rn n/2 2n n+2 2(n − 1) n−2 n Z 2+n 2 V+ kφk22 Proof. In order to prove a lower bound we may of course replace V by V+ . We use the Sobolev Inequality and Hölder’s inequality −2 2n 4 1 2(n − 1) n+2 kφk Q(φ) ≥ kφk22n − kV+ k 2+n kφk n+2 . 2n 2 2 n−2 n−2 2 n−2 We get a lower bound by minimizing over t = kφk22n , i.e., n−2 ( ) −2 4 n 1 2(n − 1) Q(φ) ≥ min t − kV+ k 2+n kφk2n+2 t n+2 , 2 t≥0 2 n−2 which gives the answer above. 8 Talenti, G. Best constant in Sobolev inequality, Ann. Mat. Pura Appl. 110 (1976), 353–372. See the book Analysis, AMS Graduate Studies in Mathematics Vol. 114 by Lieb and Loss for a simple proof. 29 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 For n = 3 we find 768 Q(φ) ≥ − 25 r Z Z 6 5/2 5/2 2 V+ kφk2 ≈ −33.65 V+ kφk22 . 5 5.11 PROBLEM. Show that using the sharp Sobolev inequality (14) we get r Z Z 27π 2 2 5/2 5/2 2 (15) V+ kφk22 . V+ kφk2 ≈ −6.74 Q(φ) ≥ − 25 5 5.12 PROBLEM (Positivity of Schrödinger quadratic form). Show that if V+ ∈ L3/2 (R3 ) and if the norm kV+ k3/2 is small enough then Z Z 1 2 Q(φ) = |∇φ| − V |φ|2 ≥ 0 2 R3 R3 for all φ ∈ C01 (R3 ). Correction ∞ n 5.13 PROBLEM. Show that if n ≥ 3 and V = V1 + V2 , where V1 ∈ L (R ) and V2 ∈ Ln/2 (Rn ) with kV2 kn/2 small enough then the closed quadratic form defined in Theorem 4.6 corresponding to the operator − 21 ∆ − V defined originally on C02 (Rn ) has domain H 1 (Rn ). [Hint: You may use that H 1 (Rn ) is the domain of the closed quadratic form in the case V = 0] 5.14 EXAMPLE (Sobolev lower bound on hydrogen). We now use the Sobolev inequality to give a lower bound on the hydrogen quadratic form Z Z 1 2 Q(φ) = |∇φ| − Z|x|−1 |φ|2 . 2 R3 3 R In Corollary 5.6 we of course already found the sharp lower bound for the hydrogen energy. This example serves more as a test of the applicability of the Sobolev inequality. For all R > 0 1 Q(φ) ≥ 2 Z 2 Z |∇φ| − R3 −1 2 Z|x| |φ| − ZR −1 |x|≤R Z |φ|2 . R3 Using (15) we find " Q(φ) ≥ 27π 2 − 25 #Z r Z 2 Z 5/2 |x|−5/2 − ZR−1 |φ|2 5 |x|<R R3 ≈ −169.43Z 5/2 R1/2 − ZR−1 ≥ −57.87Z 2 , where we have minimized over R. This result should be compared with the sharp value −0.5Z 2 . since May 3: 3 → n 30 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 We may think of the Sobolev lower bound Theorem 5.10 as a bound on the first min-max value µ1 (independetly of whether it is an eigenvalue or not) of − 21 ∆ − V (x). The Sobolev bound may be strengthened to the following result. A proof may be found in Lieb and Loss, Analysis, Theorem 12.4. 5.15 THEOREM (Lieb-Thirring inequality). There exists a constant CLT > 0 such that if V ∈ L2loc (Rn ) and V+ ∈ L(n+2)/2 (Rn ) then the min-max values µn for − 12 ∆ − V (x) defined on C02 (Rn ) satisfy ∞ X Z µn ≥ −CLT (n+2)/2 V+ . n=0 5.16 PROBLEM (Hardy’s Inequality). Show that for all φ ∈ C01 (R3 ) we have Z Z 1 2 |∇φ(x)| dx ≥ |x|−2 |φ(x)|2 dx. 4 3 3 R R Show also that 1/4 is the sharp constant in this inequality. 5.17 EXAMPLE (Dirichlet and Neumann Boundary conditions). Consider the quadratic form Z Q(φ) = 1 |φ0 |2 0 2 on the Hilbert space L ([0, 1]) with the domain D0 (Q) = φ ∈ C 1 ([0, 1]) : φ(0) = φ(1) = 0 or D1 (Q) = C 1 ([0, 1]). The operator A0 corresponding to Q according to Problem 3.9 with domain D0 (Q) satisfies that Correction since May 3: = D(A0 )∩C 2 ([0, 1]) φ ∈ C 2 ([0, 1]) : φ(0) = φ(1) = 0 = D(A0 ) ∩ C 2 ([0, 1]) added and if φ ∈ C 2 ([0, 1]) with φ(0) = φ(1) = 0 then A0 φ = −φ00 . This follows by integration by parts since if ψ ∈ D0 (Q) then Z 1 Z 0 0 0 0 Q(ψ, φ) = ψ φ = φ (1)ψ(1) − φ (0)ψ(0) − 0 0 1 00 ψφ = − Z ψφ00 . 31 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 The condition φ(0) = φ(1) = 0 is called the Dirichlet boundary condition. The operator A1 corresponding to Q with domain D1 (Q) satisfies that Correction since May 3: added and if φ ∈ C 2 ([0, 1]) with φ0 (0) = φ0 (1) = 0 then A1 φ = −φ00 . Note that this time there was no boundary condition in the domain D1 (Q), but it appeared in the domain of A1 . The boundary condition φ0 (0) = φ0 (1) = 0 is called the Neumann boundary condition. The statement again follows by integration by parts as above. This time the boundary terms do not vanish automatically. Since the map (ψ, φ) 7→ φ0 (1)ψ(1) is not bounded on L2 we have to ensure the vanishing of the boundary terms in the definition of the domain of A1 . 2 d The operators A1 and A0 are both extensions of the operator A = − dx 2 defined on D(A) = C02 (0, 1), i.e., the C 2 -functions with compact support inside the open interval (0, 1). Both A1 and A0 are bounded below and thus have Friedrichs’ extensions. We shall see below that these Friedrichs’ extensions are not the same. We will also see that the Friedrichs’ extension of A is the same as the Friedrichs’ extension of A0 . 5.18 PROBLEM. Show that the eigenvalues of the Dirichlet operator A0 in Example 5.17 are n2 π 2 , n = 1, 2, . . .. Show that the eigenvalues of the Neumann operator A1 in Example 5.17 are n2 π 2 , n = 0, 1, 2, . . .. Argue that these operators cannot have the same Friedrichs’ extension. 2 d 5.19 PROBLEM. Show that if we consider the operator A = − dx 2 defined on C02 (0, 1) then the min-max values of A are µn = n2 π 2 n = 1, 2, . . .. Hence these are eigenvalues of the Friedrichs’ extension of A. Argue that therefore AF = A0F . Show however that they are not eigenvalues for the closure of the operator A. (From the theory of Fourier series it is known that the eigenfunctions corresponding to the eigenvalues n2 π 2 n = 1, 2, . . . form an orthonormal basis for L2 (0, 1). You may assume this fact.) = D(A1 )∩C 2 ([0, 1]) φ ∈ C 2 ([0, 1]) : φ0 (0) = φ0 (1) = 0 = D(A1 ) ∩ C 2 ([0, 1]) 32 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 6 The canonical and grand canonical picture and the Fock spaces We return to the study of the N -body operator HN = HNin X + Wij = N X j=1 1≤i<j≤N X hj + Wij (16) 1≤i<j≤N defined on the Hilbert space HN = h1 ⊗ · · · ⊗ hN . This situation where we study a fixed number of particles N is referred to as the canonical picture. If we have an infinite sequence of particles (and hence also an infinite sequence of spaces h1 , h2 , . . .) we may however consider all particle numbers at the same time. To do this we introduce the Fock Hilbert space ∞ M F= h1 ⊗ · · · ⊗ hN (17) N =0 (when N = 0 we interpret h1 ⊗ · · · ⊗ hN as simply C and refer to it as the 0particle space, the vector 1 ∈ C is often called the vacuum vector and is denoted Ω or |Ωi) and the operator ∞ M H= HN , Correction H N =0 ∞ M ΨN = N =0 ∞ M since May 3: removed HN ΨN (18) N =0 (here H0 = 0) with domain ) ! ( ∞ ∞ ∞ X M M 2 kHN ΨN k < ∞ . ΨN | ΨN ∈ D(HN ), D HN = Ψ = N =0 N =0 N =0 This situation when all particle numbers are considered at the same time is called the grand canonical picture. 6.1 PROBLEM. What is the natural quadratic form domain for L∞ N =0 HN ? 6.2 DEFINITION (Stability of first and second kind). A many-body system is said to be stable of the first kind or canonically stable if the operators HN are stable for all N , i.e., if they are bounded below. A many-body system is said to be stable of the second kind or grand canonically stable if there exists a constant µ such that the operator ∞ M HN + µN N =0 (with the same domain as L∞ N =0 HN ) is stable, i.e., bounded below. = 33 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Of special interest is the situation when we have identical particles. In this case we may introduce the bosonic Fock space F B (h) = ∞ O N M h (19) h. (20) N =0 sym and the fermionic Fock space F F (h) = ∞ ^ N M N =0 The projections P± defined in (5) may be identified with projections on F with P+ (F) = F B (h), P− (F) = F F (h). In this case we refer to h as the one-particle space. 6.3 PROBLEM. Assume we have two one-paticle spaces h1 and h2 . In this problem we shall see that the spaces F B,F (h1 ⊕ h2 ) may in a natural way be identified with F B,F (h1 ) ⊗ F B,F (h2 ). More precisely, show that there is a unique unitary map U : F B (h1 ) ⊗ F B (h2 ) → F B (h1 ⊕ h2 ) such that s U (Φ1 ⊗ Φ2 ) = (N1 + N2 )! P+ (Φ1 ⊗ Φ2 ) N1 !N2 ! for Φ1 belonging to the N1 -particle sector of F B (h1 ) and Φ2 belonging to the N2 particle sector of F B (h2 ). The corresponding result holds for the fermionic Fock spaces. L B 6.4 PROBLEM. Consider an operator of the form H = ∞ N =0 HN on F, F or L F F . If Ψ is a normalized vector in the domain D ( ∞ N =0 HN ) show that there exist N and a normalized vector ΨN ∈ D(HN ) such that (Ψ, HΨ) ≥ (ΨN , HN ΨN ). If we have a system that is stable of the second kind, such that L∞ N =0 HN +µN is stable, it follows from the above problem that the corresponding ground state energy is always attained at a fixed particle number. The grand canonical picture is useful in situations where we look for the particle number which gives the smallest possible energy. JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 34 6.5 EXAMPLE (Molecules). We will here give the quantum mechanical description of a molecule. We first consider the canonical picture where the molecule has N electrons (mass= 1, charge= −1, and spin= 1/2) and K nuclei with charges Z1 , . . . , ZK > 0, masses M1 , . . . , MK , and spins j1 , . . . , jK (satisfying 2jk + 1 ∈ N for k = 1, . . . , K). Some nuclei may be identical, but let us for simplicity not treat them as bosons or fermions. The Hilbert space describing the molecule is HN = N ^ L2 (R3 ; C2 ) ⊗ K O L2 (R3 ; C2jk +1 ). k=1 The Hamiltonian is HN = N X i=1 − K X X 1 1 1 − ∆xi + − ∆rk + 2 2Mk |xi − xj | 1≤i<j≤N k=1 N X K X i=1 k=1 X Zk Z` Zk + . |xi − rk | 1≤k<`≤K |rk − r` | Here we have written the nuclei coordinates as rk ∈ R3 , k = 1, . . . , K and the electron coordinates as xi ∈ R3 , i = 1, . . . , N . We may choose the domain for HN to be functions in C02 , i.e., smooth functions Correction May 3: with compact support. It can be proved that the molecule is stable in the sense and added that there is a ground state energy EN > −∞ (See Problem A.6.1) We may now consider the grand canonical picture for the electrons, i.e., we vary the number N of electrons but leave the number K of nuclei fixed. Thus L∞ L∞ we consider the operator N =0 HN on the Fock space N =0 HN . It can be proved that this system is stable of the second kind (even with µ = 0), i.e., that inf N EN > −∞. In fact, there is an Nc such that EN = ENc for all N ≥ Nc . It is known that Z1 + . . . + ZK ≤ Nc ≤ 2(Z1 + . . . + ZK ) + 1. 6.6 PROBLEM. What would the Hilbert space be in the previous example if the nuclei were all identical bosons? 6.7 PROBLEM (Very difficult). Show that the map N 7→ EN defined in the previous example is a non-increasing map. since ∞ → 3 problem JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 35 6.8 EXAMPLE (Matter). In the previous example we considered the number of nuclei fixed, but both the canonical and the grand canonical picture for the electrons. We may also consider the grand canonical situation for the nuclei. Let us assume that we have only a finite number L of different kinds of nuclei and let us still treat them neither as bosons nor fermions. Thus we want to consider an arbitrary number K of nuclei with charges, masses and spins belonging to the set {(Z1 , M1 , j1 ), . . . , (ZL , ML , jL )} We have to specify how many of each kind of nuclei we have. Let us not do this explicitly, but only say that as the number of nuclei K tends to infinity we want to have that the fractions of each kind converge to some values ν1 , . . . , νL > 0 where of course ν1 + . . . + νL = 1. Let EN,K be the canonical ground state energy for N electrons and K nuclei. Stability of the second kind for this system states that there is a constant µ such that EN,K ≥ µ(N + K). (21) This inequality is true. It is called Stability of Matter. It was first proved by Dyson and Lenard in 1967–689,10 but has a long history in mathematical physics. Moreover, it is true that the limit inf N EN,K K→∞ K exists. This is a version of what is called the existence of the thermodynamic lim limit. It was proved in a somewhat different form by Lieb and Lebowitz11 . 7 Second quantization We now introduce operators on the bosonic and fermionic Fock spaces that are an important tool in studying many body problems. 9 Dyson, Freeman J. and Lenard, Andrew, Stability of matter. I, Jour. Math. Phys. 8, 423–434, (1967). 10 Dyson, Freeman J. and Lenard, Andrew, Stability of matter. II, Jour. Math. Phys. 9, 698–711 (1968). 11 Lieb, Elliott H. and Lebowitz, Joel L. , The constitution of matter: Existence of thermodynamics for systems composed of electrons and nuclei. Advances in Math. 9, 316–398 (1972). 36 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 For any vector in the one-particle Hilbert space f ∈ h. We first introduce two operators a(f ) and a∗ (f ) on the Fock Hilbert space F= ∞ M HN , HN = N O h. N =0 These operators are defined by the following actions on the pure tensor products a(f )(f1 ⊗ · · · ⊗ fN ) = N 1/2 (f, f1 )h f2 ⊗ · · · ⊗ fN a∗ (f )(f1 ⊗ · · · ⊗ fN ) = (N + 1)1/2 f ⊗ f1 ⊗ · · · ⊗ fN . On the 0-particle space C they act as a(f )Ω = 0 and a∗ (f )Ω = f . We extend the LM action of a(f ) and a∗ (f ) by linearity to the domain ∪∞ M =0 N =0 HN . Then a(f ) and a∗ (f ) are densely defined operators in F with the property that they map a(f ) : HN → HN −1 , a∗ (f ) : HN → HN +1 . We call a(f ) an annihilation operator and a∗ (f ) a creation operator. We think of a(f ) as annihilating a particle in the one-particle state f and of a∗ (f ) as creating a particle in this state. 7.1 PROBLEM. Show that the operators a(f ) and a∗ (f ) may be extended to the domain {Ψ = ∞ M ΨN | N =0 ∞ X N kΨN k2 < ∞}. N =0 7.2 PROBLEM. Show that for all vectors Ψ, Φ ∈ ∪∞ M =0 LM N =0 HN we have (a(f )Ψ, Φ)F = (Ψ, a∗ (f )Φ)F , when f ∈ h. For this reason we say that a(f ) and a∗ (f ) are formal adjoints. It is more important to define creation and annihilation operators on the bosonic and fermionic Fock spaces. The annihilation operators may simply be restricted to the bosonic and fermionic subspaces . The creation operators however require that we project back onto the appropriate subspaces using the projections P± defined in (5) now considered on the Fock space. Thus we define a± (f ) = a(f ), a∗± (f ) = P± a∗ (f ) (22) 37 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 7.3 PROBLEM. Show that a± (f ) and a∗± (f ) for all vectors f ∈ h define densely defined operators on the spaces F B (h) (in the + case) and F F (h) (in the − case). Show moreover that on their domains these operators satisfy (a+ (f )Ψ, Φ)F B = (Ψ, a∗+ (f )Φ)F B , (a− (f )Ψ, Φ)F F = (Ψ, a∗− (f )Φ)F F . The maps h 3 f → a∗± (f ) are linear whereas the maps h 3 f → a± (f ) are anti-linear. 7.4 PROBLEM. We introduce the commutator [A, B] = AB −BA and the anticommutator {A, B} = AB + BA of two operators. Show that on their domain of definition the operators a+ and a∗+ satisfy the Canonical Commutation Relations (CCR) [a+ (f ), a+ (g)] = [a∗+ (f ), a∗+ (g)] = 0, [a+ (f ), a∗+ (g)] = (f, g)h I (23) Show that on their domain of definition the operators a− and a∗− satisfy the Canonical Anti-Commutation Relations (CAR) {a− (f ), a− (g)} = {a∗− (f ), a∗− (g)} = 0, {a− (f ), a∗− (g)} = (f, g)h I. (24) 7.5 PROBLEM. Show that if dim h = n then dim(F F (h)) = 2n . If e1 , . . . , en are orthonormal basis vectors in h describe the action of the operators a− (ei ), a∗− (ei ) on an appropriate basis in F F (h). 7.6 PROBLEM. In this exercise we will give two descriptions of the Fock space F B (C). Show that F B (C) in a natural way may be identified with the space `2 (N) such that the vacuum vector Ω is the sequence (1, 0, 0, 0 . . .). Let |ni denote the sequence with 1 in the n-th position and 0 elsewhere. Write the actions of the operators a+ (1), a∗+ (1) on the basis vector |ni. Correction Show that we may also identify F B (C) with the space L2 (R) such that the vacuum vector Ω is the function (π)−1/4 e−x √1 (x 2 − d ). dx 2 /2 and a+ (1) = √1 (x 2 + d ), dx a∗+ (1) = For this last question it is useful to know that the space of functions of the form p(x)e−x 2 /2 , where p(x) is a polynomial, is a dense subspace in L2 (R). 1. Show that if u ∈ h is a unit vector then we have a direct L∞ sum decomposition F B (h) = n=0 Hn such that Hn is an eigenspace of 7.7 PROBLEM. eigenvalue n for the operator a∗+ (u)a+ (u). since May vacuum normalization corrected 3 vector 38 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 2. If u1 , . . . , ur ∈ h are orthonormal vectors then we have a direct sum decomL∞ L∞ position F B (h) = n1 =0 . . . nr =0 Hn1 ,...,nr such that Hn1 ,...,nr is a joint eigenspace for the operators a∗+ (u1 )a+ (u1 ),. . . , a∗+ (ur )a+ (ur ) of eigenvalues n1 , . . . , nr respectively. 3. Likewise for fermions show that if u1 , . . . , ur ∈ h are orthonormal vectors L L then we have a direct sum decomposition F F (h) = 1n1 =0 . . . 1nr =0 Hn1 ,...,nr such that Hn1 ,...,nr is a joint eigenspace for the operators a∗− (u1 )a− (u1 ),. . . , a∗− (ur )a− (ur ) with eigenvalues n1 , . . . , nr respectively. 7.8 LEMMA (2nd quantization of 1-body operator). Let h be a symmetric operator on h and let {uα }∞ α=1 be an orthonormal basis for h with elements from the domain D(h). We may then write ∞ X N M hj = N =1 j=1 ∞ X ∞ X (uα , huβ )a∗± (uα )a± (uβ ) (25) β=1 α=1 as quadratic forms on the domain ∪∞ M =0 LM N =1 P P± D( N j=1 hj ) (for M = 0 the domain is C). Proof. We first observe that if f, g ∈ D(h) then ∞ X ∞ X (g, uα )h (uα , huβ )h (uβ , f )h = β=1 α=1 ∞ X (g, huβ )h (uβ , f )h = β=1 ∞ X (hg, uβ )h (uβ , f )h β=1 = (hg, f )h = (g, hf )h . This in fact shows that the identity (25) holds in the sense of quadratic forms on D(h). Let us consider the action of a∗ (uα )a(uβ ) on a pure tensor product a∗ (uα )a(uβ )f1 ⊗ · · · ⊗ fN = N (uβ , f1 )h uα ⊗ f2 ⊗ · · · ⊗ fN where f1 , . . . , fN ∈ D(h). Thus we have ∞ X ∞ X (uα , huβ )h a∗ (uα )a(uβ ) = N h1 β=1 α=1 as quadratic forms on finite linear combinations of N -fold pure tensor products of functions from D(h). 39 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Since P± projects onto symmetrized or anti-symmetrized vectors we have P PN NN P± N h1 P± = P± N h. Hence j=1 hj P± = j=1 hj P± on ∞ X ∞ X (uα , huβ )h a∗± (uα )a± (uβ )P± = ∞ X ∞ X β=1 α=1 (uα , huβ )h P± a∗ (uα )a(uβ )P± β=1 α=1 = ∞ X N M hj P± . N =1 j=1 7.9 DEFINITION (2nd quantization of 1-body operator). The operator ∞ X N M hj N =1 j=1 is called the second quantization of the operator h. It is sometimes denoted dΓ(h), but we will not use this notation here. 7.10 REMARK. If U is an operator on h another way to lift U to the Fock space F B,F (h) is muliplicatively Γ(U ) = ∞ O N M U N =0 N ( N U = I when N = 0.) This is also denoted the second quantization of U . It is the relevant operation for transformation operators, e.g., unitary maps. consisting of vectors from a given dense subspace. Note that the second quantization of the identity operator I on h is the L B F number operator N = ∞ N =0 N on F (h) or F (h). The number operator may N = ∞ X a∗± (uα )a± (uα ), α=1 for any orthonormal basis {uα }∞ α=1 on h. We have a similar result for 2-body potentials. since May Case 3: N = 0 added 7.11 PROBLEM. Show that one can always find an orthonormal basis for h be written as Correction (26) 40 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 7.12 LEMMA (2nd quantization of 2-body operator). Let {uα }∞ α=1 be an orthonormal basis for h. Let W be a 2-body potential for identical particles, i.e., a symmetric operator on h ⊗ h such that ExW Ex = W . Assume that uα ⊗ uβ ∈ D(W ) for all α, β = 1, 2, . . .. Then as quadratic forms on finite linear combinations of pure symmetric (+) or antisymmetric (-) tensor products of basis vectors from {uα }∞ α=1 we have ∞ M X Wij = ∞ X 1 2 N =2 1≤i<j≤N (uα ⊗ uβ , W uµ ⊗ uν )a∗± (uα )a∗± (uβ )a± (uν )a± (uµ ). α,β,µ,ν=1 (27) Proof. We have W u 1 ⊗ u2 = ∞ X (uα ⊗ uβ , W uµ ⊗ uν )h⊗h (uµ , u1 )h (uν , u2 )h uα ⊗ uβ α,β,µ,ν=1 and a∗ (uα )a∗ (uβ )a(uν )a(uµ )u1 ⊗ u2 ⊗ · · · ⊗ uN = N (N − 1)(uµ , u1 )(uν , u2 )uα ⊗ uβ ⊗ u3 ⊗ · · · ⊗ uN . Since this is true for any N -fold pure tensor product of basis vectors we conclude that on such tensor products 1 2 ∞ X (uα ⊗ uβ , W uµ ⊗ uν )a∗ (uα )a∗ (uβ )a(uν )a(uµ ) = α,β,µ,ν=1 N (N − 1) W12 . 2 As in the previous proof we have that on N -fold tensor products X N (N − 1) P± W12 P± = Wij P± 2 1≤i<j≤N (we are here using that ExW Ex = W ). Note that P± a∗ (f )P± = P± a∗ (f ) (if we symmetrize or anti-symmetrize after creating an extra particle it plays no role whether we had symmetrized or anti-symmetrized before). Thus ∞ M X ∞ M N (N − 1) Wij P± = P± W12 P± 2 N =2 1≤i<j≤N N =2 = P± 12 ∞ X α,β,µ,ν=1 (uα ⊗ uβ , W uµ ⊗ uν )a∗ (uα )a∗ (uβ )a(uν )a(uµ )P± 41 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 = = 1 2 1 2 ∞ X α,β,µ,ν=1 ∞ X (uα ⊗ uβ , W uµ ⊗ uν )P± a∗ (uα )P± a∗ (uβ )a(uν )a(uµ )P± (uα ⊗ uβ , W uµ ⊗ uν )a∗± (uα )a∗± (uβ )a± (uν )a± (uµ )P± . α,β,µ,ν=1 7.13 DEFINITION (2nd quantization of 2-body operator). The operator ∞ M X Wij N =2 1≤i<j≤N is called the second quantization of the two-body operator W . 8 One- and two-particle density matrices for bosonic or fermionic states 8.1 DEFINITION (One-particle density matrix). Let Ψ = L∞ N =0 ΨN be a normalized vector on the bosonic Fock space F B (h) or the fermionic Fock space F F (h) with finite particle expectation (Ψ, N Ψ) = ∞ X N kΨN k2 < ∞. N =0 We define the 1-particle density matrix (or 2-point function) of Ψ as the operator γΨ on the one-body space h given by (f, γΨ g)h = (Ψ, a∗± (g)a± (f )Ψ). 8.2 PROBLEM. Show that γΨ is a positive semi-definite trace class operator with TrγΨ = (Ψ, N Ψ). 8.3 PROBLEM. Show that if Ψ is a finite linear combination of pure tensor products of elements from a subspace X ⊆ h then γΨ is a finite rank operator whose range is a subspace of X. Correction May 3: since Formula for Ψ corrected JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 42 8.4 THEOREM (Fermionic 1-particle density matrix). If Ψ is a normalized vector on the fermionic Fock space F F (h) then γΨ satisfies the operator inequality 0 ≤ γΨ ≤ I. (28) In particular, the eigenvalues of γΨ are in the interval [0, 1]. Proof. We simply have to prove that for all f ∈ h we have 0 ≤ (f, γΨ f )h ≤ kf k2 . The first inequality follows from Problem 7.3 since (f, γΨ f )h = (Ψ, a∗− (f )a− (f )Ψ) = (a− (f )Ψ, a− (f )Ψ) = ka− (f )Ψk2 ≥ 0. The second inequality above follows from Problem 7.3 and the CAR relations (24) since (f, γΨ f )h = (Ψ, a∗− (f )a− (f )Ψ) ≤ (Ψ, a∗− (f )a− (f )Ψ) + (a∗− (f )Ψ, a∗− (f )Ψ) = (Ψ, a∗− (f )a− (f ) + a− (f )a∗− (f )Ψ) = (Ψ, {a− (f ), a∗− (f )}Ψ) = kf k2 . 8.5 PROBLEM. If u1 , . . . , uN are orthonormal vectors in h we consider the normalized (see Problem 1.24) vector Ψ = u1 ∧ · · · ∧ uN . Show that the corresponding 1-particle density matrix γΨ is the projection in h onto the N -dimensional space spanned by u1 , . . . , uN . We are now ready to prove the result corresponding to Corollary 2.10 for fermions. 8.6 THEOREM (Ground state of non-interacting Fermi system). We consider P the Hamiltonian HNin = N j=1 hj for N identical particles restricted to the antiVN symmetric subspace h, i.e., with domain D− (HNin ) = P− D(HNin ). We assume that the one-body operator h is bounded from below. The ground state energy of this fermionic system is inf{Tr(P h) | P an orth. projection onto an N-dimensional subspace of D(h)}. (29) 43 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 If the infimum is attained for an N -dimensional projection P then HNin has as ground state eigenvector f1 ∧ · · · ∧ fN , where f1 , . . . , fN is an orthonormal basis for the space P (h). This basis may be chosen to consist of eigenvectors of h. All expectations of h restricted to the orthogonal complement P (h)⊥ ∩ D(h) will be greater than all expectations of h restricted to P (h). Notice that according to Problem 4.13 the ground state energy of the N P particle fermionic system may be described as N n=1 µn (h), where µn (h) are the min-max values of h. Proof. Choose an orthonormal basis {uα }∞ α=1 for h with vectors from D(h) (see Problem 7.11) Let Ψ ∈ D− (HNin ) be normalized. It follows from Lemma 7.8 that (Ψ, HNin Ψ) = ∞ X ∞ X (uα , huβ )(Ψ, a∗− (uα )a− (uβ )Ψ) = β=1 α=1 ∞ X ∞ X (uα , huβ )(uβ , γΨ uα ). β=1 α=1 Recall that Ψ is assumed to be a finite linear combination of pure tensor products of elements from D(h). Thus from Problem 8.3 we know that γΨ has finite rank. Let γΨ have eigenvectors v1 , . . . , vn and corresponding eigenvalues λ1 , . . . , λn . It follows again from Problem 8.3 that v1 , . . . , vn ∈ D(h). Then (Ψ, HNin Ψ) = n X ∞ X ∞ X λj (uα , huβ )(uβ , vj )(vj , uα ) = j=1 β=1 α=1 n X λj (vj , hvj ) j=1 The state Ψ has fixed particle number N therefore we have that n X λj = TrγΨ = (Ψ, N Ψ) = N. j=1 Since 0 ≤ λj ≤ 1 we must have n ≥ N . We may assume that we had chosen the eigenvectors ordered such that (v1 , hv1 ) ≤ (v2 , hv2 ) ≤ . . . ≤ (vn , hvn ). If we define the N -dimensional projection P that projects onto the space spanned by v1 , . . . , vN we find that Tr[P h] = N X j=1 (vj , hvj ) ≤ n X j=1 λj (vj , hvj ) ≤ (Ψ, HNin Ψ). JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 44 Thus inf{Tr(P h) | P an orth. projection onto an N-dimensional subspace of D(h)} ≤ inf{(Ψ, HNin Ψ) | Ψ ∈ D(HNin ), kΨk = 1} The opposite inequality is also true. In fact, given N orthonormal vectors f1 , . . . , fN ∈ D(h). Let Ψ = f1 ∧ · · · ∧ fN . Then according to Problem 8.5 γΨ = P is the projection onto the space spanned by f1 , . . . , fN . As above we find that (Ψ, HNin Ψ) = Tr[P h]. Assume now that P minimizes the variational problem in (29). It is clear from the above proof that the vector Ψ = f1 ∧ · · · ∧ fN is a ground state for HNin if f1 , . . . , fN is an orthonormal basis for P (h). We now show the stated properties of the space P (h) corresponding to a minimizing projection. It is first of all clear that if φ ∈ P (h) and ψ ∈ P (h)⊥ ∩ D(h) are normalized then (φ, hφ) ≤ (ψ, hψ). In fact, consider the projection Q onto the N -dimensional space span(P (h) ∩ {φ}⊥ ) ∪ {ψ}, i.e., the space where we have replaced φ by ψ. We then have since P is minimizing 0 ≤ Tr[Qh] − Tr[P h] = (ψ, hψ) − (φ, hφ). The same argument actually shows that if ψ ∈ D(h) ∩ (P (h) ∩ {φ}⊥ )⊥ is normalized then (φ, hφ) ≤ (ψ, hψ). We will now use this to show that h maps the space P (h) into itself. Assume otherwise, that there is a φ ∈ P (h) such that hφ 6∈ P (h). Since D(h) is dense there is a g ∈ D(h) such that ((I − P )g, hφ) = (g, (I − P )hφ) 6= 0. Assume that Re((I − P )g, hφ) 6= 0 (otherwise we simply multiply g by i). We have (I − P )g ∈ D(h). Consider for t ∈ R (close to 0) φt = φ + t(I − P )g . kφ + t(I − P )gk Note that φt ∈ D(h)∩(P (h)∩{φ}⊥ )⊥ . Hence (φ, hφ) ≤ (φt , hφt ). This is however in contradiction with the fact that d (φt , hφt )|t=0 = ((I − P )g, hφ) + (φ, h(I − P )g) = 2Re((I − P )g, hφ) 6= 0. dt 45 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Thus h maps P (h) to itself and this space is hence spanned by eigenvectors of h. 8.7 PROBLEM (Bosonic 1-particle density matrix). If φ ∈ h is normalized we consider the N -fold tensor product of φ with itself Ψ = φ ⊗ φ ⊗ · · · φ. Note that N B Ψ∈ N sym h ⊆ F (h). Determine the 1-particle density matrix γΨ . 8.1 Two-particle density matrices 8.8 DEFINITION (Two-particle density matrix). Let Ψ = L∞ N =0 be a normal- B ized vector on the bosonic Fock space F (h) or the fermionic Fock space F F (h) with 2 (Ψ, N Ψ) = ∞ X N 2 kΨN k2 < ∞. N =0 We define the 2-particle density matrix (or 4-point function) of Ψ as the operator (2) ΓΨ on the two-body space h ⊗ h uniquely given by (2) (f1 ⊗ f2 , ΓΨ g1 ⊗ g2 )h⊗h = (Ψ, a∗± (g2 )a∗± (g1 )a± (f1 )a± (f2 )Ψ). (30) (in the fermionic case the ordering of the creation and annihilation operators is important). Correction 8.9 PROBLEM. Show that (30) indeed uniquely defines a positive semi-definite (2) June equation added, trace class operator ΓΨ is with (2) Show also that (2) (2) ExΓΨ = ±ΓΨ (31) where (+) is for bosons and (−) is for fermions The exchange operator Ex was defined on Page 9. 8.10 PROBLEM. (Compare Problem 8.5) If u1 , . . . , uN are orthonormal vectors in h we consider the normalized (see Problem 1.24) vector Ψ = u1 ∧ · · · ∧ uN . (2) Show that the corresponding 2-particle density matrix ΓΨ is given by ΓΨ = γΨ ⊗ γΨ − ExγΨ ⊗ γΨ 24: number problem reformulated TrΓΨ = (Ψ, N (N − 1)Ψ). (2) since JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 46 (see Page 9 for the definition of the tensor product of operators). Determine the (2) eigenvectors and eigenvalues of ΓΨ and conclude, in particular, that the largest (2) eigenvalue of ΓΨ is at most 2. Correction since May 3 1 → 2 8.11 PROBLEM (Bosonic 2-particle density matrix). Determine the 2-particle density matrix for the bosonic state in Problem 8.7. 8.12 THEOREM (Fermionic 2-particle density matrix). If dim h = M and V Ψ ∈ N h for some N ≤ M and Ψ is normalized then if N and M are even (2) 0 ≤ ΓΨ ≤ N (M − N + 2) I. M (32) For all M including M = ∞ we have 0≤ (2) ΓΨ ≤ N I. (33) 8.13 REMARK. • For N = 2 the upper bound in (32) is equal to the simple (2) ΓΨ ≤ N (N − 1)I, which follows from Problem 8.9. For all N > 2 bound Correction since May 3: Last sentence of lemma improved the bound in (32) is strictly smaller than the simple bound. (This is left for the reader to check.) (2) • If N = M the upper bound in (32) is ΓΨ ≤ 2I. Only in this case does the upper bound in (32) agree with the upper bound for Slater determinants (see Problem 8.10). • We shall see in the proof of Theorem 8.12 that the upper bound in (32) is achieved for special pair states, in which certain pairs of states are either both occupied or both empty. This is an example of what is called Cooper pairs and states of this type was very important in the famous BardeenCooper-Schrieffer theory of superconductivity12 Correction May 3: • A discussion of other results on bounds of 1-,2- and n-point functions and since there → their their relations to physics may be found in a classical paper of C.N. Yang13 In order to prove the above theorem we need a little lemma. Correction May 3: 12 J. Bardeen, L.N. Cooper, and J.R. Schrieffer, Theory of Superconductivity, Phys. Rev., formulation corrected 108, 1175–1204 (1957). 13 C.N. Yang, Concept of Off-Diagonal Long Range Order and the Quantum Phases of Liquid He and of Superconductors, Rev. of Mod. Phys., 34, 694–704, 1962. since Lemma 47 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 8.14 LEMMA. If dim(h) = M and f ∈ h ∧ h there exist orthonormal vectors u1 , . . . , u2r , where r is a positive integer less than M/2 and λ1 , . . . , λr ≥ 0 such that f= r X λi u2i−1 ∧ u2i = λ1 u1 ∧ u2 + λ2 u3 ∧ u4 + . . . . i=1 This lemma is proved in Appendix E. Proof of Theorem 8.12. (See also Appendix A in the paper by C.N.Yang in footnote 13.) We will write M = 2m and N = 2n where m, n are positive integers. We will proceed by induction on M . If M = 2 then N must be 2 (the case N = 0 is trivial). If u1 , u2 is an orthonormal basis for M then the only possible state with two particles is Ψ = u1 ∧ u2 . This is the case studied in Problem 8.10, where we saw that indeed the largest eigenvalue is 2 = N (M −N +2) . M The same argument actually may be used whenever M = N . Assume now that M > 2 and N < M − 2 and that the theorem has been proved for M − 2 and all N ≤ M − 2. V (2) Let f ∈ h ⊗ h and Ψ ∈ N h be normalized vectors such that (f, ΓΨ f )h⊗h is as large as possible. Then similarly to Problem 2.5 we conclude that f is an (2) eigenvector of ΓΨ . As a consequence of (31) f is antisymmetric, i.e., f ∈ h ∧ h. According to Lemma 8.14 we may write f= m X λi u2i−1 ∧ u2i . i=1 Let ai = a− (ui ) hence a∗i = a∗− (ui ). Define F = m X √ 2λi a2i−1 a2i . i=1 (2) The definition of ΓΨ implies that (2) (f, ΓΨ f )h⊗h = (Ψ, F ∗ F Ψ)F F (h) . (34) Pm λ2i = 1 we may without loss of generality assume Pm √ 2λi a2i−1 a2i such that F = F 0 + that λ1 > 0. Let us introduce F 0 = i=2 √ 2λ1 a1 a2 . Then Since f is normalized, i.e., F ∗ F = F 0∗ F 0 + i=1 √ 2λ1 F 0∗ a1 a2 + √ 2λ1 a∗2 a∗1 F 0 + 2λ21 a∗2 a∗1 a1 a2 (35) 48 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 We write Ψ = Φ00 + Φ01 + Φ10 + Φ11 , corresponding to the direct sum decomposition described in Problem 7.7(3) for the operators a∗1 a1 and a∗2 a2 , i.e., Φk` for k, ` = 0, 1 is the projection of Ψ onto the subspace where the number of particles in states u1 and u2 are k and ` or more explicitly a∗1 a1 Φk` = kΦk` , and a∗2 a2 Φk` = `Φk` . From (35) we obtain (Ψ, F ∗ F Ψ) = (Φ00 , F 0∗ F 0 Φ00 ) + (Φ01 , F 0∗ F 0 Φ01 ) + (Φ10 , F 0∗ F 0 Φ10 ) +(Φ11 , (F 0∗ F 0 + 2λ21 a∗2 a∗1 a1 a2 )Φ11 ) √ √ +(Φ00 , 2λ1 F 0∗ a1 a2 Φ11 ) + (Φ11 , 2λ1 a∗2 a∗1 F 0 Φ00 ) = (Φ00 , F 0∗ F 0 Φ00 ) + (Φ01 , F 0∗ F 0 Φ01 ) + (Φ10 , F 0∗ F 0 Φ10 ) +(Φ11 , F 0∗ F 0 Φ11 ) + 2λ21 (Φ11 , Φ11 ) √ √ + 2λ1 (Φ00 , F 0∗ a1 a2 Φ11 ) + (Φ11 , 2λ1 a∗2 a∗1 F 0 Φ00 ). We observe that this expression is a sum of two quadratic forms (Ψ, F ∗ F Ψ) = Q1 (Φ00 + Φ11 ) + Q2 (Φ01 + Φ10 ). We will now argue that without changing the value of (Ψ, F ∗ F Ψ) we may assume that either Φ00 = Φ11 = 0 or Φ01 = Φ10 = 0. In fact, if this were not already the case we would have ∗ (Ψ, F F Ψ) ≤ max Q1 (Φ00 + Φ11 ) Q2 (Φ01 + Φ10 ) , kΦ00 k2 + kΦ11 k2 kΦ01 k2 + kΦ10 k2 , i.e., we would do just as well by choosing either Φ00 = Φ11 = 0 or Φ01 = Φ10 = 0. Correction since May 3: 01 → 10 Since Q2 does not depend on λ1 we see that the best choice cannot be Φ00 = Φ11 = 0 because in this case we could increase the value of Q2 (Φ01 + Φ10 ) to P (1 − λ21 )−1 Q2 (Φ01 + Φ10 ) by replacing f by f 0 = (1 − λ21 )−1/2 ri=2 λi u2i−1 ∧ u2i in contradiction with the fact that the value was already chosen optimal. Therefore Correction since we may assume that Φ01 = Φ10 = 0. We therefore have (Ψ, F ∗ F Ψ) = (Φ00 , F 0∗ F 0 Φ00 ) + (Φ11 , F 0∗ F 0 Φ11 ) + 2λ21 (Φ11 , Φ11 ) √ √ + 2λ1 (Φ00 , F 0∗ a1 a2 Φ11 ) + (Φ11 , 2λ1 a∗2 a∗1 F 0 Φ00 ). May 3: Explanation of Φ01 = Φ10 = 0 improved JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 49 We will now apply the Cauchy-Schwarz inequality (Φ00 , F 0∗ a1 a2 Φ11 ) + (Φ11 , a∗2 a∗1 F 0 Φ00 ) = (F 0 Φ00 , a1 a2 Φ11 ) + (a1 a2 Φ11 , F 0 Φ00 ) ≤ 2(a1 a2 Φ11 , a1 a2 Φ11 )1/2 (Φ00 , F 0∗ F 0 Φ00 )1/2 ≤ 2(Φ11 , a∗2 a∗1 a1 a2 Φ11 )1/2 (Φ00 , F 0∗ F 0 Φ00 )1/2 = 2(Φ11 , Φ11 )1/2 (Φ00 , F 0∗ F 0 Φ00 )1/2 . Inserting this into the identity above we finally obtain (Ψ, F ∗ F Ψ) ≤ (Φ00 , F 0∗ F 0 Φ00 ) + (Φ11 , F 0∗ F 0 Φ11 ) √ +2λ21 (Φ11 , Φ11 ) + 2 2λ1 (Φ11 , Φ11 )1/2 (Φ00 , F 0∗ F 0 Φ00 )1/2 . (36) Since Φ00 ∈ VN h0 , where h0 =span{u3 , . . . , uM } and F 0 only contains a3 , . . . , aM we infer from the induction hypothesis that N (M − N ) N (M − N ) = (1 − λ21 )(1 − kΦ11 k2 ) M −2 M −2 V = u1 ∧ u2 ∧ Φ0 , with Φ0 ∈ N −2 h0 we get (Φ00 , F 0∗ F 0 Φ00 ) ≤ (1 − λ21 )kΦ00 k2 Likewise since Φ11 (Φ11 , F 0∗ F 0 Φ11 ) ≤ (1 − λ21 )kΦ11 k2 (N − 2)(M − N + 2) . M −2 If we denote by Y = kΦ11 k we have 0 ≤ Y ≤ 1. We conclude from (36) that (Ψ, F ∗ F Ψ) ≤ G(λ, Y ) where N (M − N ) M −2 (N − 2)(M − N + 2) +(1 − λ21 )Y 2 + 2λ21 Y 2 M −2 1/2 √ N (M − N ) 2 1/2 2 1/2 +2 2λ1 Y (1 − λ1 ) (1 − Y ) . M −2 G(λ, Y ) = (1 − λ21 )(1 − Y 2 ) (37) One may now analyze (see Appendix D for details) the function G(λ, Y ) and show that for 0 ≤ λ ≤ 1 and 0 ≤ Y ≤ 1 its maximum is attained for λ1 = (2/M )1/2 , where the maximal value is N (M −N +2) . N Y = (N/M )1/2 . (38) 50 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 It follows from the proof above that the upper bound in Theorem 8.12 is optimal and one may go through the proof to find the Ψ that optimizes the bound. In the next example we directly construct a Ψ that achieves the bound in Theorem 8.12. 8.15 EXAMPLE (A 2-particle density matrix with maximal eigenvalue). Our goal in this example is to show that the bound in the previous example is, in fact, V optimal. We will explicitly write down a state ΨN ∈ N h, where dim h = M (2) such that the 2-particle density matrix ΓΨN has eigenvalue N (M −N +2) , M i.e., the largest possible. We will assume that M = 2m and N = 2n where m and n are positive integers. Let u1 , . . . , uM be an orthonormal basis for h. Let ai = a− (ui ). Hence a∗i = a∗− (ui ). We define first a vector that does not have a fixed particle number m Y Ψ = c0 (1 + a∗2j−1 a∗2j )|0i. (39) j=1 Here c0 is a positive normalization constant. We choose ΨN to be a normalized V vector in N h proportional (by a positive constant) to the projection of Ψ onto VN h. We have n Y −1/2 X m ΨN = n 1≤i <...<i 1 We claim that f= m X n ≤m a∗2ik −1 a∗2ik |0i. k=1 m−1/2 u2i−1 ∧ u2i i=1 (2) is an eigenfunction of ΓΨN with the largest possible eigenvalue. 8.16 PROBLEM. Show that the above vector f is normalized. Let F = m X √ 2m−1/2 a2i−1 a2i . i=1 Then as in (34) we have (2) (f, ΓΨN f ) = (ΨN , F ∗ F ΨN ). 8.17 PROBLEM. Show (combinatorically) that (ΨN , F ∗ F ΨN ) = N (M −N +2) . M 51 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 (2) It follows from the previous problem that (f, ΓΨN f ) has the largest possible (2) value among normalized f . It follows as in Problem 2.5 (used on −ΓΨN ) that f (2) must be an eigenvector of ΓΨN with eigenvalue N (M −N +2) . M The state in (39) as well as the Slater determinant states in Problem 1.24 and 8.5 are special cases of what are called quasi-free states. We will study these states in more detail in Section 10. 8.2 Generalized one-particle density matrix If Ψ ∈ F B,F (h) does not have a fixed particle number it is also important to know (Ψ, a± (f )a± (g)Ψ) and (Ψ, a∗± (f )a∗± (g)Ψ). We will therefore consider a generalization of the one-particle density matrix. 8.18 DEFINITION (Generalized one-particle density matrix). If Ψ ∈ F B,F (h) is a normalized vector we define the corresponding generalized one-particle density matrix to be the positive semi-definite operator ΓΨ defined on h ⊕ h∗ by (f1 ⊕ Jg1 , ΓΨ f2 ⊕ Jg2 )h⊕h∗ = (Ψ, (a∗± (f2 ) + a± (g2 ))(a± (f1 ) + a∗± (g1 ))Ψ)F B,F . Here as usual J : h → h∗ is the conjugate linear map such that Jg(f ) = (g, f ). 8.19 PROBLEM. Show that ΓΨ as defined above is indeed linear. 8.20 PROBLEM. Show that for fermions 0 ≤ ΓΨ ≤ Ih⊕h∗ . (Compare Theorem 8.4). We may write ΓΨ in block matrix form corresponding to the direct sum decomposition h ⊕ h∗ ΓΨ = γΨ αΨ ∗ αΨ 1 ± JγΨ J ∗ ! . (40) Here + is for bosons and − is for fermions. The map γΨ : h → h is the usual one-particle density matrix and αΨ : h∗ → h is the linear map given by (f, αΨ Jg) = (Ψ, a± (g)a± (f )Ψ). (41) The adjoint J ∗ of the conjugate linear map J, which is, in fact, also the inverse J ∗ = J −1 , is discussed in Appendix E. The fact that the lower right corner of 52 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 the matrix has the form given follows from the canonical commutation or anticommutation relations in Problem 7.4 (Ψ, a± (g2 )a∗± (g1 )Ψ) = (g2 , g1 ) ± (Ψ, a∗± (g1 )a± (g2 )Ψ) = (g2 , g1 ) ± (g2 , γΨ g1 ) = (Jg1 , Jg2 ) ± (Jg1 , JγΨ J ∗ Jg2 ) = (Jg1 , (1 ± JγΨ J ∗ )Jg2 ). The linear map αΨ has the property ∗ αΨ = ±JαΨ J. (42) In fact, from the definition (41) we have ∗ f, Jg) = (f, αΨ Jg) = ±(g, αΨ Jf ) = ±(JαΨ Jf, Jg). (αΨ Thus we may also write the generalized one-particle density matrix as ! γΨ αΨ ΓΨ = , ±JαΨ J 1 ± JγΨ J ∗ (43) where + is for bosons and − is for fermions. It will also be convenient to introduce the generalized annihilation and creation operators A± (f ⊕ Jg) = a± (f ) + a∗± (g) (44) A∗± (f ⊕ Jg) = a∗± (f ) + a± (g). (45) Note that A± is a conjugate linear map from h ⊕ h∗ to operators on F B,F (h) and A∗± is a linear map. We have the relation A∗± (F ) = A± (J F ) where J = 0 J∗ J ! : h ⊕ h ∗ → h ⊕ h∗ , 0 (46) for all F ∈ h ⊕ h∗ . Using the generalized creation and annihilation operators we may express the canonical commutation relations as [A+ (F1 ), A∗+ (F2 )] = (F1 , SF2 )h⊕h∗ , where S = 1 0 0 −1 ! : h⊕h∗ → h⊕h∗ (47) and the canonical anti-commutation relations as {A− (F1 ), A∗− (F2 )} = (F1 , F2 )h⊕h∗ . (48) JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 53 We warn the reader that in general [A+ (F1 ), A+ (F2 )] 6= 0, {A− (F1 ), A− (F2 )} = 6 0. In terms of the generalized creation and annihilation operators the generalized one-particle density matrix satisfies (F1 , ΓΨ F2 )h⊕h∗ = (Ψ, A± (F2 )∗ A± (F1 )Ψ), (49) for all F1 , F2 ∈ h ⊕ h∗ . 9 Bogolubov transformations 9.1 DEFINITION (Bogolubov maps). A linear bounded isomorphism V : h ⊕ h∗ → h ⊕ h∗ is called a bosonic Bogolubov map if A∗+ (VF ) = A+ (VJ F ) for all F ∈ h ⊕ h∗ and [A+ (VF1 ), A∗+ (VF2 )] = (F1 , SF2 ), (50) for all F1 , F2 ∈ h ⊕ h∗ . It is called a fermionic Bogolubov map if A∗− (VF ) = A− (VJ F ) for all F ∈ h ⊕ h∗ and {A− (VF1 ), A∗− (VF2 )} = (F1 , F2 ). (51) for all F1 , F2 ∈ h ⊕ h∗ . This definition simply says that a Bogolubov map V is characterized by F 7→ A± (VF ) having the same properties ((46) and (47) or (48)) as F 7→ A± (F ). If V is a Bogolubov map then one often refers to the operator transformation A± (F ) → A± (VF ) as a Bogolubov or (Bogolubov-Valatin) transformation. Using (46) and (47) we may rewrite the conditions for being a bosonic Bogolubov map as (VF1 , SVF2 ) = (F1 , SF2 ), and J VF = VJ F for all F, F1 , F2 ∈ h ⊕ h∗ . Likewise, using (46) and (48) we may rewrite the conditions for being a fermionic Bogolubov map as (VF1 , VF2 ) = (F1 , F2 ), and J VF = VJ F JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 54 for all F, F1 , F2 ∈ h ⊕ h∗ . Since we are assuming that V is invertible we see that V −1 = SV ∗ S (52) V −1 = V ∗ (53) in the bosonic case and in the fermionic case. Thus we immediately conclude the following reformulation of the definition of Bogolubov maps. Correction since June 24: Theorem 9.2 THEOREM (Bogolubov maps). A linear map V : h ⊕ h∗ → h ⊕ h∗ is a bosonic Bogolubov map if and only if V ∗ SV = S, VSV ∗ = S, J VJ = V. (54) It is a fermionic Bogolubov map if and only if V ∗ V = Ih⊕h∗ , VV ∗ = Ih⊕h∗ , J VJ = V. (55) A fermionic Bogolubov map is, in particular, unitary. 9.3 PROBLEM. Show that the Bogolubov maps form a subgroup of the group of isomorphism of h ⊕ h∗ . We may write a Bogolubov map as a block matrix ! U J ∗V J ∗ V= , V JU J ∗ (56) where U : h → h, V : h → h∗ . That a Bogolubov map must have the special matrix form (56) follows immediately from the condition J VJ = V. In order for the matrix (56) to be a Bogolubov map we see from (54) and (55) that U and V must also satisfy the conditions U ∗ U ∓ V ∗ V = 1, JV ∗ JU ∓ JU ∗ J ∗ V = 0 (57) where − is for bosons and + is for fermions. We also get from (54) and (55) that U U ∗ ∓ J ∗V V ∗J = 1 (58) reformulated JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 55 again with − for bosons and + for fermions. We shall next show that the Bogolubov transformations A± (F ) 7→ A± (VF ) may be implemented by a unitary map on the Fock spaces F B,F . We will need the following result. 9.4 PROBLEM. Assume that u1 , u2 . . . are orthonormal vectors in h. Consider for some positive integers, M, n1 , . . . , nM the vector a∗± (uM )nM · · · a∗± (u1 )n1 |0i ∈ F B,F (h). Show that in the bosonic case the vector has norm (n1 ! · · · nM !)1/2 for all nonnegative integers n1 , . . . , nM . Show that in the fermionic case the vector vanishes unless n1 , . . . , nM are all either 1 and in this cases the vector is normalized. 9.5 THEOREM (Unitary Bogolubov implementation). If V : h ⊕ h∗ → h ⊕ h∗ is a Bogolubov map (either fermionic or bosonic) of the form (56) then there exists a unitary transformation UV : F B,F (h) → F B,F (h) such that UV A± (F )U∗V = A± (VF ) for all F ∈ h ⊕ h∗ if and only if V ∗ V is trace class. This trace class condition is referred to as the Shale-Stinespring condition. Proof. The proof is somewhat complicated and we will give a sketchy presentation leaving details to the interested reader. We assume first V ∗ V is trace class and will construct the unitary UV . Let u1 , u2 , . . . be an orthonormal basis for h. We have an orthonormal basis for F B,F (h) given by (see Problem 9.4) |ni1 , . . . , niM i = (ni1 ! · · · niM !)−1/2 a∗± (uiM )niM · · · a∗± (ui1 )ni1 |0i, where M = 1, 2, . . ., 1 ≤ i1 < i2 < . . . < iM run over the positive integers. For bosons n1 , . . . , nM run over positive integers and for fermions they are all 1. We will construct the unitary UV by constructing the orthonormal basis |ni1 , . . . , niM iV = UV |ni1 , . . . , niM i. JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 56 The main difficulty is to construct the vacuum |0iV = UV |0i. Recall that A± (u ⊕ 0) = a± (u) are the annihilation operators and that A± (0 ⊕ Ju)) = a∗± (u) are the creation operators. Thus if UV exists the new vacuum must be characterized by A± (V(ui ⊕ 0))|0iV = UV A± (ui ⊕ 0)U∗V UV |0i = UV A± (ui ⊕ 0)|0i = 0, for all i = 1, 2, i.e., by being annihilated by all the new annihilation operators A± (V(ui ⊕ 0)). We shall construct the new vacuum below. Having constructed the new vacuum |0iV the rest of the proof is fairly easy. We must have |ni1 , . . . , niM iV = UV |ni1 , . . . , niM i = (ni1 ! · · · niM !)−1/2 UV A± (0 ⊕ JuiM )niM · · · A± (0 ⊕ Jui1 )ni1 U∗V UV |0i = (ni1 ! · · · niM !)−1/2 A± (V(0 ⊕ JuiM ))niM · · · A± (V(0 ⊕ Jui1 ))ni1 |0iV . It follows from the fact, that the new creation operators and the new annihilation operators satisfy the canonical commutation or anti-commutation relations that the vectors |ni1 , . . . , niM iV will form an orthonormal family. All we have to show is that they form a basis. To do this we simply revert the construction and construct the old basis vectors |ni1 , . . . , niM iV from the new |ni1 , . . . , niM iV by simply interchanging the roles of the old and the new basis vectors and of V and V −1 . Doing this we will be able to express the old basis vectors as (possibly infinite) linear combinations of the new basis vectors, thus showing that the new vectors indeed span the whole space. We will leave this reversion of the construction to the interested reader. It remains to construct the new vacuum. We first choose a particularly useful orthonormal basis of h. We use the notation of (56). Note that the linear Hermitian matrix U ∗ U commutes with the conjugate linear map C = U ∗ J ∗ V . In fact, from (58) and (57) we have U ∗U C = U ∗U U ∗J ∗V = U ∗ (1 ± J ∗ V V ∗ J)J ∗ V = U ∗ J ∗ V ± U ∗ J ∗ V V ∗ V = U ∗ J ∗ V + U ∗ J ∗ V (U ∗ U − 1) = CU ∗ U. Since V ∗ V is trace class it has an orthonormal basis of eigenvectors. The relation (57) shows that this is also an eigenbasis for U ∗ U . It follows from (58) that C ∗ C = V ∗ JU U ∗ JV = V ∗ (1 ± V V ∗ )V = V ∗ V ± (V ∗ V )2 57 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 and thus C ∗ C is trace class. Since the eigenvalues of U ∗ U are real it follows that C maps each eigenspace of U ∗ U into itself. Indeed, if v ∈ h satisfies U ∗ U v = λv for λ ∈ R we have U ∗ U Cv = CU ∗ U v = λCv. From (57) we see that the map C is a conjugate Hermitian map for bosons and a conjugate anti-Hermitian map for fermions. We may therefore find an orthonormal basis for each eigenspace of U ∗ U according to Theorem E.2. This means that we can find an orthonormal basis u1 , u2 . . . of h consisting of eigenvectors of U ∗ U , denoting the eigenvalues µ21 , µ22 . . . (assuming that µ1 , . . . ≥ 0), such that in the bosonic case they are also eigenvectors of C with real eigenvalues λ1 , λ2 . . ., or in the fermionic case such that they fall in two groups ui , i ∈ I 0 and ui , i ∈ I 00 with I 0 ∪ I 00 = N, I 0 ∩ I 00 = ∅ such that Cu2i = λi u2i−1 , Cu2i−1 = −λi u2i , 2i ∈ I 0 where λi > 0 and Cui = 0, i ∈ I 00 . We have according to (56) the new annihilation operators A± (V(ui ⊕ 0)) = A± ((U ui ⊕ V ui )) = µi a± (fi ) + a∗± (gi ), (59) where for i = 1, 2 . . . we have introduced gi = J ∗ V ui and ( µ−1 if µi 6= 0 i U ui , fi = . 0, if µi = 0 The new creation operators are (of course) the adjoints of the annihilation operators, but this indeed agrees with (56) since A± (V(0 ⊕ Jui )) = A± ((J ∗ V ui ⊕ JU ui ) = a± (gi ) + µi a∗± (fi ). In the bosonic case it follows from (57) that µi ≥ 1, thus in this case we ∗ have (fi , fj ) = µ−2 i (ui , U U uj ) = δij and the fi are orthonormal. Since U is ∗ a surjective map (since U is invertible by (58)) the fi form an orthonormal −1 ∗ basis for h. Moreover, (fi , gj ) = µ−1 i (U ui , J V uj ) = µi (ui , Cuj ) = λi /µi δij . Let νi = λi /µi . Thus gi = νi fi . From (57) νi2 = λ2i = (gi , gi ) = (ui , V ∗ V ui ) = (ui , (U ∗ U − 1)ui ) = µ2i − 1. µ2i Correction since June 22 injective → surjective 58 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 We conclude that in the bosonic case there is an orthonormal basis f1 , f2 . . . , for h and numbers µi ≥ 1, νi ∈ R, for i = 1, . . . such that the new bosonic annihilation operators are A+ (V(ui ⊕ 0)) = µi a+ (fi ) + νi a∗+ (fi ), µ2i − νi2 = 1 (60) for i = 1, 2 . . .. We can now in the bosonic case find the new vacuum vector |0iV characterized by being annihilated by all the new annihilation operators. Indeed, |0iV n ∗ ∞ M X Y a+ (fj )2n −νj 2 1/4 (1 − (νj /µj ) ) = lim |0i M →∞ 2µ n! j n=0 j=1 n ∗ ∞ Y X a+ (fj )2n −νj 2 1/4 = |0i (1 − (νj /µj ) ) 2µ n! j n=0 j=1 ! " # Y X νi 2 1/4 ∗ 2 = (1 − (νj /µj ) ) exp − a (fi ) |0i. 2µi + j=1 i=1 (61) Here the exponential is really just a convenient way of writing the power series. The normalization factor follows from the Taylor series expansion 2 −1/2 (1 − 4t ) = ∞ 2n X t (2n)! n=0 (n!)2 , which gives for all i 2 1/2 (1 − (νi /µi ) ) ∞ X h0|a+ (fi )2n a∗+ (fi )2n |0i(νi /2µi )2n (n!)2 n=0 2 1/2 = (1 − (νi /µi ) ) ∞ X (2n)!(νi /2µi )2n (n!)2 n=0 Using that V ∗ V is trace class and hence that P i=1 = 1. νi2 < ∞ we shall now see that the limit M → ∞ above exists. Define ΨM " M # M Y X νi = (1 − (νj /µj )2 )1/4 exp − a∗+ (fi )2 |0i. 2µ i j=1 i=1 We have 2 kΨN − ΨM k = 2 − 2 N Y j=M +1 (1 − (νj /µj )2 )1/4 → 0 59 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 as M → ∞ uniformly in N > M . Thus ΨM is Cauchy sequence. Since fi is orthogonal to fj for i 6= j the creation and annihilation operators a+ (fi ), a∗+ (fi ) commute with a+ (fj ), a∗+ (fj ) if i 6= j. Using this it is a fairly straightforward calculation to see that (µi a+ (fi ) + νi a∗+ (fi ))|0iV = 0 for all i = 1, . . ., which is what we wanted to prove. We turn to the fermionic case. Our goal is to show that if we define ( fi , if µi 6= 0 ηi = gi , if µi = 0 (62) then η1 , η2 . . . is an orthonormal basis for h. We claim moreover that the new annihilation operators may be written Correction since June 24: i → 2i A− (V(η2i−1 ⊕ 0)) = αi a− (η2i−1 ) − βi a∗− (η2i ), 2i ∈ I 0 (63) A− (V(η2i ⊕ 0)) = αi a− (η2i ) + βi a∗− (η2i−1 ), 2i ∈ I 0 (64) A− (V(ηi ⊕ 0)) = a∗− (ηi ), i ∈ Ik00 (65) A− (V(ηi ⊕ 0)) = a− (ηi ), i ∈ I 00 \ Ik00 , (66) where k is a non/negative integer and Ik00 rfers to the first k elements of I 00 , αi = µ2i , βi ≥ 0 and αi2 + βi2 = 1, for 2i ∈ I 0 . Before proving this we observe that it is easy to see from this representation that the following normalized vector is annihilated by all the operators in (63–65) Y Y |0iV = a∗− (ηi ) (αi + βi a∗− (η2i )a∗− (η2i−1 ))|0i i∈Ik00 = 2i∈I 0 Y i∈Ik00 a∗− (ηi ) ! Y 2i∈I 0 αi ! X βi a∗− (η2i )a∗− (η2i−1 ) |0i. (67) exp α i 2i∈I 0 Again this vector should really be defined by a limiting procedure. Since V ∗ V P Q is trace class we have from (57) that i (1 − µ2i ) < ∞ and thus 2i∈I 0 αi2 = Q 2 2i∈I 0 µ2i . The limiting procedure is hence justified as in the bosonic case. To prove (63)–(65) we return to (59). We have as in the bosonic case (fi , fj ) = −1 µ−1 i µj (U ui , U uj ) = δij if µi , µj 6= 0. From (57) we have (gi , gj ) = (uj , V ∗ V ui ) = (uj , (1 − U ∗ U )ui ) = (1 − µ2i )δij . several places JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 60 Thus for all i, j = 1, 2, . . . with i 6= j, gi is orthogonal to gj and if µi = 0 gi is normalized. If µi 6= 0 and µj = 0 we have, since U ∗ U and C commute, that −1 ∗ (fi , gj ) = µ−1 i (U ui , J V uj ) = µi (ui , Cuj ) −3 ∗ ∗ = µ−3 i (U U ui , Cuj ) = µi (ui , CU U uj ) = 0. Hence η1 , η2 , . . . defined in (62) is an orthonormal family in h. We will show that it is a basis. It is clear that the vectors ηi for i with µi = 0 pan Ran(U ). The orthogonal complement is Ker(U ∗ ). If g ∈ Ker(U ∗ ) it follows from (58) that J ∗ V V ∗ Jg = g. If we can show that U V ∗ Jg = 0 then V ∗ Jg is spanned by ui corrsponding to µi = 0. Hence g is in the span of the gi corresponding to such i, i.e., the case when gi = ηi . That U V ∗ Jg = 0 follows from U ∗ U V ∗ Jg = (1 − V ∗ V )V ∗ Jg = V ∗ Jg − V ∗ JJ ∗ V V ∗ Jg = V ∗ Jg − V ∗ Jg = 0. If for some 2i ∈ I 0 , µ2i = 0 we have U u2i = 0 and hence λi u2i−1 = Cu2i = −C ∗ u2i = V ∗ JU u2i = 0, but this is a contradiction with λi > 0. Thus for all 2i ∈ I 0 we have µ2i 6= 0 and likewise µ2i−1 6= 0. Thus ηi = fi for 2i ∈ I 0 . Moreover, for 2i ∈ I 0 and all j we have (U uj , g2i ) = (U uj , J ∗ V u2i ) = (uj , Cu2i ) = λi δ2i−1,j . (68) Thus since g2i is orthogonal to all gj with j 6= 2i we conclude that g2i = λi µ−1 2i−1 f2i−1 . (69) g2i−1 = −λi µ−1 2i f2i . (70) and likewise We know that for 2i ∈ I 0 ∗ ∗ 2 λ2i µ−2 2i−1 = (g2i , g2i ) = (V V u2i , u2i ) = ((1 − U U )u2i , u2i ) = 1 − µ2i and likewise that 2 λ2i µ−2 2i = 1 − µ2i−1 . These two identities imply that µ2i = µ2i−1 . JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 61 Then (63–64) follow from (69)–(70) with αi = µ2i = µ2i−1 and βi = λi µ−1 2i = λi µ−1 2i−1 . For i ∈ I 00 we have (gi , U uj ) = (J ∗ V ui , U uj ) = (Cui , uj ) = 0. Thus gi is orthogonal to all fj . Since gi is also orthogonal to all gj with i 6= j we conclude that either gi = 0 and hence µi = 1 or gi = ηi and hence µi = 0. In the former case we must have A− (V(ui ⊕ 0)) = a− (ηi ) and in the latter case A− (V(ui ⊕ 0)) = a∗− (ηi ). Since V ∗ V is trace class the eigenvalue 1 has finite multiplicity. This means that the eigenvalue µi = 0 for U ∗ U has finite multiplicity k. We can assume that I 00 has been ordered such that µi = 0 occurs for the first k i. The necessity of the Shale-Stinespring condition is proved in Appendix F. In the rest of this chapter we will for simplicity assume that the space h is finite dimensional. 9.6 LEMMA. Assume dimh = M and that a Hermitian A : h ⊕ h∗ → h ⊕ h∗ satisfies J AJ = ±A (+ for bosons and − for fermions) (71) and moreover in the Bose case that A is positive definite. Then there exists a bosonic (+) or fermionic (-) Bogolubov map V such that the operator V ∗ AV has eigenvectors of the form u1 ⊕0, . . . , uM ⊕0, 0⊕Ju1 , . . . , 0⊕JuM , where u1 , . . . , uM is an orthonormal basis for h. Proof. We will construct V by finding the vectors v1 = V(u1 ⊕ 0), . . . , v2 = V(uM ⊕ 0), vM +1 = V(0 ⊕ Ju1 ), . . . , v2M = V(0 ⊕ JuM ). We first consider the fermionic case. It is straightforward to check that V will satisfy the required properties if v1 , . . . , v2M form an orthonormal basis of eigenvectors of A such that for all j = 1, . . . , M vM +j = J vj . JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 62 Let v1 be a normalized eigenvector of A with eigenvalue λ1 . Define vM +1 = J v1 . Then vM +1 is a normalized vector and from (71) we have that AvM +1 = −J Av1 = −J λ1 v1 = −λ1 vM +1 , where we have used that λ1 is real. Thus vM +1 is an eigenvector of A. Moreover, if λ1 6= 0 it follows that the eigenvalues λ1 and −λ1 are different and hence that vM +1 is orthogonal to v1 . We may then restrict A to the orthogonal complement of the space spanned by v1 and vM +1 and continue the process in this way we will find an orthonormal family of vectors of the desired form. They will however not necessarily form a basis since we still have to consider the kernel of A. This eigenspace must be even dimensional since the whole space is even dimensional and we have just constructed a basis for the orthogonal complement consisting of an even number of vectors. It follows from (71) that J maps the kernel of A to itself. We may then using Theorem E.2 find an orthonormal basis for the kernel consisting of eigenvectors of J with non-negative eigenvalues. Since, J 2 = I the eigenvalues are 1. If w1 and w2 are two basis vectors the vectors v± = w1 ± iw2 are orthonormal and they satisfy J v± = v∓ . By pairing the basis vectors for the kernel of A in this manner we find a basis of the desired form. This completes the proof in the fermionic case. We turn to the bosonic case. It is again straightforward to check that we have to show the existence of a basis v1 , . . . , v2M for h ⊕ h∗ with the following properties 1. (vi , Svj ) = ±δij , with + for i = 1, . . . , M and − for i = M + 1, . . . , 2M . 2. For all j = 1, . . . , 2M we have SAvj = λj vj for some λj ∈ C. 3. J vj = vj+M for all J = 1, . . . , M . Note that item 2 is not an eigenvalue problem for A, but for SA, which is not Hermitian. Still it can be analyzed in much the same way as the eigenvalue problem for a Hermitian matrix. 63 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 First of all the values λj have to be real since both A and S are Hermitian and Avj = λj Svj thus (vj , Avj ) = λj (vj , Svj ). (72) Any family of vectors satisfying item 1 is linearly independent, as is easily seen, in the same way as one concludes that orthonormal vectors are linearly independent. Thus if we can find v1 , . . . v2M satisfying 1–3 it will automatically be a basis. The eigenvalues in 2 are of course roots in the characteristic polynomial det(SA − λI) = 0. Since this equation has at least one root we can find v1 and a corresponding value λ1 satisfying 2. Since A is positive definite we see from (72) that λ1 6= 0 and (v1 , Sv1 ) 6= 0. We can therefore assume that v1 has been normalized in such a way that (v1 , Sv1 ) = 1 Define vM +1 = J v1 then using (71) we have that SAvM +1 = SJ Av1 = SJ λ1 v1 = −λ1 SvM +1 , where we have used that λ1 is real and that J S = −SJ . Thus vM +1 satisfies 2 with λM +1 = −λ1 . Since λ1 6= 0 then λM +1 6= λ1 and we conclude from λM +1 (v1 , SvM +1 ) = (v1 , AvM +1 ) = (Av1 , vM +1 ) = λ1 (v1 , SvM +1 ) that (v1 , SvM +1 ) = 0 and v1 and vM +1 . Since we also have (vM +1 , SvM +1 ) = (J v1 , SJ v1 ) = (J v1 , −J Sv1 ) = −(Sv1 , v1 ) = −(v1 , Sv1 ) = −1 we see that v1 and vM +1 satisfy item 1. We next show that SA maps the subspace X = {w | (v1 , Sw) = (vM +1 , Sw) = 0} into itself. Indeed, if w is in this space we have (v1 , SSAw) = (v1 , Aw) = (Av1 , w) = λ1 (Sv1 , w) = 0 and likewise for vM +1 . Hence SA must have an eigenvector on X and we can continue the argument by induction. JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 64 9.7 PROBLEM. Consider h = C. The map J may be identified with complex conjugation. We use the basis 1 ⊕ 0 and 0 ⊕ 1 for h ⊕ h∗ . In this basis we consider ! 1 a A= , a 1 for 0 < a < 1. Show that we have a bosonic Bogolubov map defined by √ 1−a2 −1) ( a √√ √√ 1−a2 −1+a2 1−a2 −1+a2 1 V=√ √ 1−a2 −1) ( 2 a √√ √√ 1−a2 −1+a2 Correction 1−a2 −1+a2 which diagonalizes A and determine the diagonal elements. What happens when a = 1? 9.8 THEOREM (Diagonalizing generalized 1pdm). Let Ψ ∈ F B,F (h) be a normalized vector and u1 , . . . , uM be an orthonormal basis for h. Then there exists a Bogolubov map V : h ⊕ h∗ → h ⊕ h∗ , such that the corresponding unitary map UV has the property that the generalized 1-particle density matrix of U∗V Ψ is a diagonal matrix in the orthonormal basis u1 ⊕ 0, . . . , uM ⊕ 0, 0 ⊕ Ju1 , . . . , 0 ⊕ JuM for h ⊕ h∗ . Proof. Using (49) it is straightforward to check that ΓU∗V Ψ = V ∗ ΓΨ V. We observe that (43) may be reformulated as J (ΓΨ + 21 S)J = ΓΨ + 12 S J (ΓΨ − 12 I)J = 1 I 2 − ΓΨ for bosons (73) for fermions.. (74) In the bosonic case we also have (f + ⊕Jg, (ΓΨ + 21 S)f ⊕ Jg)h⊕f h∗ = (Ψ, (a∗+ (f ) + a+ (g))(a+ (f ) + a∗+ (g))Ψ) +(Ψ, (a∗+ (g) + a+ (f ))(a+ (g) + a∗+ (f ))Ψ) ≥ 0. Hence ΓΨ + 12 S ≥ 0. In fact, this operator is not only positive semi-definite it is positive definite. To show this we must show that the kernel of ΓΨ + 21 S is trivial. The relation (73) shows that J leaves invariant this kernel. since June 24: b→B JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 65 From Theorem E.2 we see that there must be an element G in the kernel such that J G = λG where λ ≥ 0. Moreover, since J 2 = I we must have λ = 1 or λ = −1. We have thus shown that there must be an element of the form G = f ⊕ Jf in the kernel of ΓΨ + 21 S. Thus (Ψ, (a∗+ (f ) + a+ (f ))(a+ (f ) + a∗+ (f ))Ψ) = (G, (ΓΨ + 21 S)G) = 0. Hence Ψ is in the kernel of a∗+ (f ) + a+ (f ). It is left to the reader to show that if f 6= 0 then the kernel of a∗+ (f )+a+ (f ) is trivial (see Problem 9.9). The statement of the theorem now follows from Lemma 9.6. 9.9 PROBLEM. Show that if f ∈ h and f 6= 0 then the kernel of a∗+ (f ) + a+ (f ) is trivial. 10 Quasi-free states 10.1 DEFINITION (Quasi-free pure states). A vector Ψ ∈ F F,B (h) is called a quasi-free pure state if there exists a Bogolubov map V : h ⊕ h∗ → h ⊕ h∗ which is unitarily implementable on F F,B (h) such that Ψ = UV |0i, where UV : F F,B (h) → F F,B (h) is the unitary implementation of V. 10.2 THEOREM (Wick’s Theorem). If Ψ ∈ F F,B (h) is a quasi-free pure state and F1 , . . . , F2m ∈ h ⊕ h∗ for m ≥ 1 then Correction since June 24: π → σ (Ψ, A± (F1 ) · · · A± (F2m )Ψ) = (75) X (±1)σ (Ψ, A± (Fσ(1) )A± (Fσ(2) )Ψ) · · · (Ψ, A± (Fσ(2m−1) )A± (Fσ(2m) )Ψ), σ∈P2m and (Ψ, A± (F1 ) · · · A± (F2m−1 )Ψ) = 0. Here P2m is the set of pairings P2m = {σ ∈ S2m | σ(2j − 1) < σ(2j + 1), j = 1, . . . , m − 1, σ(2j − 1) < σ(2j), j = 1, . . . , m}. Note that the number of pairings is (2m)! . 2m m! (76) 66 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 10.3 PROBLEM. Prove Theorem 10.2. According to Theorem 10.2 we can calculate all expectations of quasi-free pure states from knowing only the generalized 1-particle density matrix. Recall, in fact, that (F1 , ΓΨ F2 ) = (Ψ, A∗± (F2 )A± (F1 )Ψ) = h0|A∗± (VF2 )A± (VF1 )|0i. In particular, we have that the expected particle number of a quasi-free pure state is (Ψ, N Ψ) = X = X = X (Ψ, a∗ (fi )a(fi )Ψ) i=1 h0|(a∗± (U fi ) + a± (V fi ))(a± (U fi ) + a∗± (V fi ))|0i i=1 h0|(a± (V fi )a∗± (V fi )|0i = i=1 X (V fi , V fi ) = TrV ∗ V, i=1 where (fi ) is an orthonormal basis for h. We see that the expected particle number is finite since we assume that V satisfies the Shale-Stinespring condition. In the next theorem we characterize the generalized 1-particle density matrices of quasi-free pure states. 10.4 THEOREM (Generalized 1-pdm of quasi-free pure state). If Ψ ∈ F B,F (h) is a quasi-free pure state then the generalized 1-particle density matrix Γ = ΓΨ : h ⊕ h∗ → h ⊕ h∗ satisfies For fermions: Γ is a projection, i.e., Γ2 = Γ For bosons: ΓSΓ = −Γ. (77) Conversely, if Γ : h ⊕ h∗ → h ⊕ h∗ is a positive semi-definite operator satisfying (77) of the form Γ= γ α ±JαJ 1 ± JγJ ∗ ! . (78) with γ a trace class operator, then there is a quasi-free pure state Ψ ∈ F B,F (h) such that ΓΨ = Γ. 67 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Proof. Since Ψ is a quasi-free pure state we may assume that Ψ = UV |0i for a Bogolubov map V : h ⊕ h∗ → h ⊕ h∗ . Thus for all F1 , F2 ∈ h ⊕ h∗ we have according to Theorem 9.5 (F1 , V ∗ ΓΨ VF2 ) = (VF1 , ΓΨ VF2 ) = (Ψ, A∗± (VF2 )A± (VF1 )Ψ) = h0|A∗± (F2 )A± (F1 )|0i. If we write Fi = fi ⊕ Jgi , i = 1, 2 we have h0|A∗± (F2 )A± (F1 )|0i = h0|(a∗± (f2 ) + a± (g2 ))(a± (f1 ) + a∗± (g1 )|0i = (g2 , g1 )h . We conclude that 0 0 ∗ V ΓΨ V = 0 I ! . From (54) or (55) we find that 0 0 ΓΨ = S± VS± 0 I ! S± V ∗ S± = S± V 0 0 0 I ! V ∗ S± , where we have introduced the notation S− = I and S+ = S. Hence using (54) or (55) we find ΓΨ S± ΓΨ = S± V 0 0 0 I ! S± 0 0 ! 0 I V ∗ S± = ∓S± V ! 0 0 V ∗ S± = ∓ΓΨ . 0 I To prove the converse assume now that Γ is a positive semi-definite Hermitian operator satisfying (77) and of the form (78). Then γ 2 ∓ αα∗ = ∓γ, α∗ γ = JγJ ∗ α∗ , and α∗ = ±JαJ. Define V= (1 ± γ)1/2 ∓αJ(1 ± γ)−1/2 J ∗ −α∗ (1 ± γ)−1/2 J(1 ± γ)1/2 J ∗ ! . Here the operator (1 ± γ)1/2 is defined by (1 ± γ)1/2 ui = (1 ± λi )1/2 ui , using an eigenbasis u1 , u2 , . . . of the trace class operator γ, with γui = λi γi . If λi = 1 in the fermionic case we define −α∗ (1 − γ)−1/2 ui = Jui . JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 68 It is then a straightforward calculation to check that V defines a Bogolubov map and that V ∗ ΓV = 0 0 ! 0 I . Moreover, If we write V in the form (56) we see that TrV ∗ V = Tr(1 ± γ)−1/2 αα∗ (1 ± γ)−1/2 = Trγ < ∞. Hence the Shale-Stinespring condition is satisfied and we can implement V as a unitary on Fock space. It follows that Γ is the generalized one-particle density matrix of UV |0i. 11 Quadratic Hamiltonians For simplicity we assume again in this section that h is finite dimensional. 11.1 DEFINITION (Quadratic Hamiltonians). Let A : h ⊕ h∗ → h ⊕ h∗ be a Hermitian operator and assume moreover in the bosonic case that it is positive definite. The operator HA± = 2M X (Fi , AFj )A∗± (Fi )A± (Fj ), i,j=1 where F1 , . . . , F2M is an orthonormal basis for h ⊕ h∗ is called a bosonic (+) of fermionic (−) quadratic Hamiltonian corresponding to A. 11.2 PROBLEM. Show that HA± is Hermitian and is independent of the choice of basis for h ⊕ h∗ used to define it. 11.3 LEMMA. If A : h ⊕ h∗ → h ⊕ h∗ is Hermitian (and positive definite in the bosonic case) we may find a Hermitian operator A0 : h ⊕ h∗ → h ⊕ h∗ (which is positive definite in the bosonic case) satisfying J A0 J = ±A0 (+ in the bosonic case and − in the fermionic case) such that Correction June 24: since sign corrected in (79) HA±0 = HA± ± 21 Tr(AS± )I, where S+ = S and S− = I. (79) 69 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Proof. Using the CCR or CAR relations (47) and (48) we have HA± = ± 2M X (Fi , AFj )A± (Fj )A∗± (Fi ) i,j=1 = ± 2M X ∓ 2M X (Fi , AFj )(Fj , S± Fi ) i,j=1 (Fi , AFj )A∗± (J Fj )A± (J Fi ) ∓ Tr(AS± ) i,j=1 where we have also applied (46). If we now use that from (89) (Fi , AFj ) = (AFi , Fj ) = (J J AJ J Fi , Fj ) = (J Fj , J AJ J Fi ) we get HA± = ± 2M X (J Fj , J AJ J Fi )A∗± (J Fj )A± (J Fi )−Tr(AS± ) = ±HJ±AJ ∓Tr(AS± ). i,j=1 The last equality follows since J F1 , . . . , J F2M is an orthonormal basis for h ⊕ h∗ . Thus if we define A0 = 12 (A ± J AJ ) we have J A0 J = ±A0 and the relation (79) holds. We next study how quadratic Hamiltonians change under unitary implementations of Bogolubov transformations. 11.4 THEOREM (Quadratic Hamiltonians and Bogolubov unitaries). If V : h ⊕ h∗ → h ⊕ h∗ is a Bogolubov map and UV : F B,F (h) → F B,F (h) is the corresponding unitary implementation then for all Hermitian A : h ⊕ h∗ → h ⊕ h∗ we have Correction UV HA± U∗V = ± HVAV ∗. Proof. This follows from Theorem 9.5 since UV HA± U∗V = 2M X (Fi , AFj )A∗± (VFi )A± (VFj ) i,j=1 = 2M X (Fi , AFj )(Fα , VFi )A∗± (Fα )A± (Fβ )(VFj , Fβ ) i,j,α,β=1 = 2M X α,β=1 ± (V ∗ Fα , AV ∗ Fβ )A∗± (Fα )A± (Fβ ) = HVAV ∗. June 24: since V and V ∗ interchanged 70 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Combining Lemma 11.3, Theorem 11.4 and Lemma 9.6 we see that any quadratic Hamiltonian HA± is up to an additive constant unitarily equivalent to a quadratic Hamiltonian HD± with D being diagonal in a basis of the form u1 ⊕ 0, . . . , uM ⊕ 0, 0 ⊕ Ju1 , . . . , 0 ⊕ JuM . Moreover, since D = VA0 V ∗ for a Bogolubov map V we see that in the bosonic case D is positive definite because A0 was assumed positive definite. Since J A0 J = ±A0 and V commutes with J we see that J DJ = ±D. Hence if we denote the first M diagonal entries of D by d1 , . . . , dM , then the next M entries are ±d1 , . . . , ±dM . Hence HD± = = M X i=1 M X di (a∗± (ui )a± (ui ) ± a± (ui )a∗± (ui )) 2di a∗± (ui )a± (ui ) ± i=1 M X di . i=1 Observe that the operators a∗+ (ui )a+ (ui ), i = 1, 2, . . . , M commute and have nonnegative integers as eigenvalues. Thus in the bosonic case, where the diagonal P entries are positive it is clear that the ground state energy is M i=1 di and the vacuum vector |0i is a ground state eigenvector of HD+ . This means, in particular, that the original quadratic Hamiltonian HA+ has a quasi-free pure ground state eigenvector. The ground state energy of HA+ is − 21 Tr(AS) + M X di i=1 In the fermionic case we have that the operators a∗− (ui )a− (ui ), i = 1, 2, . . . , M commute and have eigenvalues 0 and 1. Thus the ground state energy of HD− is M X i=1 di ≤0 2di − M X di = − i=1 M X |di |. i=1 If we assume that the eigenvalues of D have been ordered in such a way that d1 , . . . , dM ≥ 0 then a (not necessarily unique, since some diagonal entries may vanish) ground state of HD− is given by the vacuum vector |0i. Therefore also HA− Correction since June subscript 24: M corrected has a quasi-free pure state as ground state eigenvector. Its ground state energy is Correction 1 Tr(A) 2 − M X i=1 We have proved the following result. since |di |. June 24: signed corrected in displayed equation 71 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 11.5 THEOREM (ground state eigenvector for quadratic Hamiltonian). A quadratic Hamiltonian has a ground state eigenvector which is a quasi-free pure state. 11.6 REMARK. We could also have defined a quadratic bosonic Hamiltonian HA+ if A is only positive semi-definite. In this case, however, the Hamiltonian may not have a ground state eigenvector. 11.7 PROBLEM. Let h = C2 and let a1± = a± (1, 0) and a2± = a± (0, 1). Find the ground state energy and ground state of the two Hamiltonians Correction by July 6 Hamiltonian H± = (1 + b)(a∗1± a1± + a∗2± a2± ) + b(a∗1± a∗2± + a2± a1± ) corrected where b > 0. 12 Generalized Hartree-Fock Theory Generalized Hartree-Fock theory is a theory for studying interacting fermions. In generalized Hartree-Fock theory one restricts attention to quasi-free pure states. According to Theorem 10.4 the set of all 1-particle density matrices of quasi-free fermionic pure states is14 Corrections since June 24: G HF re- G HF = Γ : h ⊕ h∗ → h ⊕ h∗ | Γ∗ = Γ has the form (43) , Γ2 = Γ, Trγ < ∞ . Let us for Γ ∈ G HF denote by ΨΓ ∈ F F (h) the (normalized) quasi-free fermionic state having Γ as its 1-particle density matrix. We consider a fermionic operator in the grand canonical picture, i.e., an operator on the Fock space F F (h) H= ∞ N M X N =0 i=1 ! hi + X Wi,j . (80) 1≤i<j≤N 12.1 DEFINITION (Generalized Hartree-Fock theory). The generalized HartreeFock functional for the operator H is map E HF : G HF → R defined by E HF (Γ) = (ΨΓ , HΨΓ ). 14 We need no longer assume that h is finite dimensional defined 72 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 The Hartree-Fock ground state energy is E HF = inf{E HF (Γ) | Γ ∈ G HF }. If E HF = E HF (Γ0 ) we call Γ0 (and ΨΓ0 ) for a Hartree-Fock ground state. There are several results in the mathematical physics literature that establish existence of a minimizer of the generalized Hartree-Fock functional. There are also several results on how well the Hartree-Fock energy approximates the true ground state energy. Since the generalized Hartree-Fock energy is found by minimizing over a restricted set of states we have the following obvious result. 12.2 THEOREM (Hartree-Fock energy upper bound on true energy). If E F = inf{(Ψ, HΨ) | Ψ ∈ F F (h), kΨk = 1} denotes the true fermionic ground state energy and E HF the Hartree-Fock ground state energy for the Hamiltonian H we have E F ≤ E HF . Hartree-Fock theory has been widely used in chemistry to calculate the energy and structure of atoms and molecules. It is fairly successful but has over the years been generalized in various ways. Using Theorem 10.2 we may calculate E HF (Γ) explicitly. Assume that Γ is written in the form (43), i.e., Γ= γ α α∗ 1 − JγJ ∗ ! , (81) It is convenient to introduce the vector α e ∈ h ⊗ h by Correction since June 24: ⊕ → ⊗ several places (f ⊗ g, α e)h⊗h = (f, αΨΓ Jg) = (ΨΓ , a− (g)a− (f )ΨΓ ). A straightforward calculation using Theorem 10.2 then shows that Correction since June Parentheses E HF (Γ) = Trh [hγ] + 21 Trh⊗h [W (γ ⊗ γ − Exγ ⊗ γ)] + 12 (e α, W α e)h⊗h . (82) included 24: JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 73 12.3 PROBLEM. Prove (82) using that if we choose an orthonormal basis u1 , u2 , . . . for h and denote a− (uα ) = aα we may write the operator H in second quantized form according to (25) and (27) as X X H= (uα ⊗ uβ , W uµ ⊗ uν )a∗α a∗β aν aµ . (uα , huβ )a∗α aβ + 21 α,β,µ,ν α,β=1 If α = 0 we say that Γ represents a normal Hartree-Fock state. In this case ΨΓ is a Slater determinant. If α 6= 0 we call the state ΨΓ a BCS state after Bardeen, Cooper and Schrieffer (see footnote 12 on Page 46) who used these type of states to explain the phenomenon of super-conductivity. We now mention without proof a result that implies that if W ≥ 0 we may restrict to normal states. 12.4 THEOREM. If W ≥ 0 then E HF = inf{E HF (Γ) | Γ ∈ G HF , Γ has the form (81) with α = 0}. It is important in the BCS theory of superconductivity that the minimizing Hartree-Fock state is not normal. For this reason it is important to understand where an attractive (negative) two-body interaction between electrons may come from. It turns out that such an attraction may be explained because of the interaction of the electrons with the atoms in the superconducting material. More precisely, it has to do with the vibrational modes that the electrons excite in the crystal of atoms. 13 Bogolubov Theory Bogolubov theory is the bosonic analogue of Hartree-Fock theory. We consider again a Hamiltonian of the form (80) but now on the bosonic Fock space F B (h). In Bogolubov theory one however does not restrict to quasi-free pure states, but to a somewhat extended class. To explain this we need a result whose proof we leave as an exercise to the reader. 13.1 THEOREM. If φ ∈ h there exists a unitary Uφ : F B (h) → F B (h) such that for all f ∈ h U∗φ a+ (f )Uφ = a+ (f ) + (f, φ). JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 74 13.2 PROBLEM. Prove Theorem 13.1. Hint: You may proceed as in the proof of Theorem 9.5 (or one may proceed from an entirely algebraic point of view). In Bogolubov theory we restrict to states of the form Uφ UV |0i, where φ ∈ h and V : h ⊕ h∗ → h ⊕ h∗ is a Bogolubov map. Another way to say this is to first perform a unitary transformation U∗φ HUφ and then to restrict to quasi-free pure states. 13.3 REMARK. We saw in Section 11 that quadratic Hamiltonians have quasifree ground states. If, in the bosonic case, we allow the quadratic operators to have terms that are linear in creation and annihilation operators then the ground states belong o the larger class of vectors for the form Uφ UV |0i. According to Theorem 10.4 the set of all 1-particle density matrices of quasifree bosonic pure states is Corrections since G Bo ∗ June 24: Definition of G Bo ∗ = {Γ : h ⊕ h → h ⊕ h | Γ has the form (43) , Γ ≥ 0, ΓSΓ = −Γ, Trγ < ∞} . changed Let us for Γ ∈ G Bo denote by ΨΓ ∈ F B (h) the (normalized) quasi-free bosonic state having Γ as its 1-particle density matrix. 13.4 DEFINITION (Bogolubov theory). The Bogolubov functional for the operator H is the map E Bo : G Bo × h → R defined by E Bo (Γ, φ) = (ΨΓ , U∗φ HUφ ΨΓ ). The Bogolubov ground state energy is E Bo = inf{E Bo (Γ, φ) | Γ ∈ G Bo , φ ∈ h}. If E Bo = E Bo (Γ0 , φ0 ) we call (Γ0 , φ0 ) (and Uφ0 ΨΓ0 ) for a Bogolubov ground state. As for the Hartree-Fock energy we also have that the Bogolubov energy is an upper bound on the true energy. 13.5 THEOREM (Bogolubov energy upper bound on true energy). If E B = inf{(Ψ, HΨ) | Ψ ∈ F B (h), kΨk = 1} denotes the true bosonic ground state energy and E Bo the Bogolubov ground state energy for the Hamiltonian H we have E B ≤ E Bo . 75 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 We leave it to the reader to use Theorem 10.2 to explicitly express the Bogolubov energy in terms of the components γ and α of Γ. 15 13.6 REMARK. In 1967 F. Dyson used Bogolubov theory to prove that a system of charged bosons does not satisfy stability of the second kind. 13.1 The Bogolubov approximation We shall finish this section by explicitly discussing an approximation introduced by Bogolubov in his study of superfluidity. We will consider bosons moving on a 3-dimensional torus of size L > 0. We identify the torus with [0, L)3 . The Hilbert space is h = L2 ([0, L)3 ). We have an orthonormal basis given by up (x) = L−3/2 exp(ipx), p ∈ 2π 3 Z. L We have the one-body operator h = −∆ − µ where −∆ is the Laplacian with periodic boundary conditions and µ > 0 is simply a parameter (the chemical potential). This means that hup = (p2 − µ)up . For the two-body potential we shall use a function that depends only on the distance between the particles. More precisely on the distance on the torus, i.e, on min |x − y − Lk|. k∈Z3 This means we have c (p − p0 )δp+q,p0 +q0 , (up ⊗ uq , W up0 ⊗ uq0 ) = L−3 W where we have introduced the Fourier coefficients of W Z c (k) = W W (x) exp(−ikx)dx. [0,L)3 15 Dyson, Freeman J., Ground state energy of a finite system of charged particles, Jour. Math. Phys. 8, 1538–1545 (1967) Dyson’s result has been improved to give the exact asymptotic energy in the large particle limit for a system of charged bosons see E.H. Lieb and J.P. Solovej, Ground State Energy of the Two-Component Charged Bose Gas. Commun. Math. Phys. 252, 485 - 534, (2004) and J.P. Solovej, Upper Bounds to the Ground State Energies of the Oneand Two-Component Charged Bose Gases. To appear in Commun. Math. Phys. 266, Number 3, 797–818 2006 Correction since June 24: Footnote changed 76 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 Let us write a+ (up ) = ap . We may then express the Hamiltonian in second quantized form as X H= (p2 − µ)a∗p ap + Z3 p∈ 2π L 1 2L3 X c (k)a∗p+k a∗q−k aq ap . W Z3 k,p,q∈ 2π L We shall now explain the Bogolubov approximation for this Hamiltonian. Let λ > 0 be a parameter. We define Hλ = U∗λu0 HUλu0 . Then (where all sums are over 2π 3 Z) L Hλ = X (p2 − µ)a∗p ap + p 1 Xc W (k)a∗p+k a∗q−k aq ap 2L3 k,p,q λ4 c λ3 c − λ µ + 3 W (0) + 3 W (0)(a0 + a∗0 ) −µλ(a0 + 2L L λ2 c X ∗ λ2 X c + 3 W (0) ap ap + 3 W (p)(a∗p ap + a∗−p a−p + a∗p a∗−p + ap a−p ) L 2L p p X λ c (p)(2a∗q+p aq ap + a∗ a∗ aq + a∗ a∗ aq ). W + 3 (83) p+q −p p q−p 2L p,q a∗0 ) 2 The motivation for using the transformation Uλu0 is that one believes that most Correction June 24: particles occupy the state u0 , called the condensate. After the transformation we look for states to restrict to the subspace where a∗0 a0 = 0. Thus λ2 is the expected number of particles in the condensate. We choose λ to minimize the two constant 4 c (0). terms above, without creation or annihilation operators, i.e., −λ2 µ + λ 3 W 2L Thus we choose λ such that µ = Hλ = X p p2 a∗p ap + λ2 c W (0). L3 Then 1 Xc λ4 c ∗ ∗ W (k)a a a a − W (0) q p p+k q−k 2L3 k,p,q 2L3 λ2 X c W (p)(a∗p ap + a∗−p a−p + a∗p a∗−p + ap a−p ) 2L3 p λ Xc + 3 W (p)(2a∗q+p aq ap + a∗p+q a∗−p aq + a∗p a∗q−p aq ). 2L p,q + (84) After the unitary transformation Uλ with the specific choice of λ one would guess that the ground state should in some sense be close to the vacuum state. Bogolubov argues therefore that one may think of the operators ap and a∗p as being small. For this reason Bogolubov prescribes that one should ignore the since Choice of λ explained JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 77 terms with 3 and 4 creation and annihilation operators. This approximation, called the Bogolubov approximation, has only been mathematically justified in a few limiting cases for specific interactions. If we nevertheless perform this approximation we arrive at the quadratic Hamiltonian X h eλ = 1 H p2 (a∗p ap + a∗−p a−p ) 2 Z3 p∈ 2π L + i λ2 c λ4 c ∗ ∗ ∗ ∗ W (p)(a a + a a + a a + a a ) − W (0). p −p p −p p −p p −p L3 2L3 (85) Since this Hamiltonian is quadratic it has a quasi-free state as ground state. This is one of the motivations why one for the original Hamiltonian restricts attention to these states. Correction since June 24: Problem 13.7 PROBLEM. (a) Show that the ground state energy of the quadratic Hamiltonian 2a∗0 a0 + a∗0 a∗0 + a0 a0 is −1. (Hint: Use the result of Problem 7.6 to √ identify a∗0 + a0 with the multiplication operator 2x on the space L2 (R).) c (p) ≥ 0 and (b) Assume that W (x) is smooth with compact support and that W c (0) > 0. Use the method described in Section 11, in particular, ProbW eλ. lem 11.7, to calculate the ground state energy of H (c) If we keep µ fixed what is then the ground state energy per volume in the limit as L → ∞? (You may leave your answer as an integral.) reformulated 100 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 A Extra Problems A.1 Problems to Section 1 A.1.1. Show that (4) defines a bounded operator K. A.1.2. (Difficult) Show that K is a compact operator on a Hilbert space if and only if it maps the closed unit ball to a compact set. Hint to the “if” part: 1. Show that there exists a normalized vector u1 that maximizes kKu1 k. 2. Show that u1 is an eigenvector for K ∗ K (see the hint to Problem 2.5) 3. By induction show that there exists an orthonormal family u1 , u2 , . . . of eigenvectors of K ∗ K that span the closure of the range of K ∗ K, which is the orthogonal complement of the kernel of K ∗ K. 4. Show that (4) holds with λn = kKun k and vn = kKun k−1 Kun (remember to check that v1 , v2 . . . is an orthonormal family) Hint to the “only if” part: Assume K can be written in the form (4). Let φ1 , φ2 , . . . be a sequence of vectors from the closed unit ball. By the BanachAlouglu Theorem for Hilbert spaces (Corollary B.2 below) there is a weakly convergent subsequence φn1 , φn2 , . . . with a weak limit point φ in the closed unit ball. Show that limk→∞ Kφnk = Kφ strongly. Conclude that the image of the closed unit ball by the map K is compact. A.1.3. Use the result of the previous problem to show that the sum of two compact operators is compact. (This is unfortunately not immediate from Definition 1.16.) A.1.4. Assume that 0 ≤ µn ≤ 1 for n = 1, 2, . . . with n = 1, 2, . . . are unit vectors in a Hilbert space H. (a) Show that the map Γ : H → H given by Γu = ∞ X n=1 µn (φn , u)φn P∞ n=1 µn = 1 and φn , 101 JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 is compact and symmetric. [Hint: Use the characterization in Problem A.1.2 (repeat the argument in the “only if” part)]. (b) Show that Γ is trace class with TrΓ = 1. A.1.5. Let σ ∈ SN and Uσ : NN H→ NN be the unitary defined in Subsec- tion 1.1. Show that Uσ Uτ = Uτ σ . A.1.6. For N ≥ 2 show that N O H⊥ N ^ H. Sym A.1.7. With the notation of Subsection 1.1 show that for all σ ∈ SN Uσ P± = (±1)σ P± . What does this mean for the action of Uσ on NN Sym H and on VN H? A.1.8. Show that if K maps bounded sequences converging weakly to zero in sequences converging strongly to 0 then K is compact. [Hint use the characterization in Problem A.1.2.] A.1.9. Show that if K is an operator on a Hilbert space such that ∞ X kKφk k2 < ∞ k=1 for some orthonormal basis {φk }∞ k=1 then K is Hilbert Schmidt. [Hint: use the result of the previous problem to show that K ∗ is compact and hence from the definition of compactness that K is compact.] A.2 Problems to Section 2 A.2.1. Show that if A is a symmetric operator on a Hilbert space then (φ, A2 φ) = (φ, Aφ)2 for some unit vector φ ∈ D(A) if and only if φ is an eigenvector of A. We interpret this as saying that a measurement of A in a given state φ always gives the same value if and only if φ is an eigenvector of A. Correction since May Two 3 problems added. JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 102 A.2.2. (Some remarks on the representation (9)) In general (9) may not make sense for a general unbounded operator A even if all ψn ∈ D(A). For simplicity we will here consider only bounded A. The general statistical average of pure states would be of the form hAi = ∞ X µn (φn , Aφn ) n=1 where 0 ≤ µn ≤ 1 for n = 1, 2, . . . with P∞ n=1 µn = 1 and φn , n = 1, 2, . . . are unit vectors, but not necessarily orthonormal. Use the result of Problem A.1.4 to show that we can find unique 0 ≤ λn ≤ 1 P for n = 1, 2, . . . with ∞ n=1 λn = 1 and ψn , n = 1, 2, . . . orthonormal such that hAi = ∞ X λn (ψn , Aψn ). n=1 A.2.3. Show that the interacting Hamiltonian HN for N identical Particles satisfies HN Uσ = Uσ HN for all permutations σ. Conclude that HN P± = P± HN and V N that HN therefore maps the subspaes N h and N SYM h into themselves. A.3 Problems to Section 3 Correction May A.3.1. Assume that K is a positive semi-definite operator defined on a Hilbert space (full domain) such that ∞ X (φk , Kφk ) < ∞ k=1 for some orthonormal basis {φk }∞ k=1 . (a) Use the Cauchy-Schwartz inequality for the quadratic form Q(φ) = (φ, Kφ) to show that for all vectors u 2 kKuk ≤ (Ku, u) ∞ X (φk , Kφk ). k=1 (b) Show that K is a bounded operator (c) Use the result of Problem A.1.9 to show that K is Hilbert-Schmidt 3 added. since Problem JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 103 (d) Show that K is trace class and that TrK = ∞ X (φk , Kφk ). k=1 A.6 Problems to Section 6 A.6.1. Show that the molecular Hamiltonian in Example 6.5 is stable. B The Banach-Alaoglu Theorem We shall here give a proof of the Banach-Alaoglu Theorem. It is one of the most useful tools from abstract functional analysis. Usually this is proved using Tychonov’s Theorem and thus relies on the axiom of choice. In the separable case this is however not necessary and we give a straightforward proof here. B.1 THEOREM (Banach-Alaoglu). Let X be a Banach space and X ∗ the dual Banach space of continuous linear functionals. Assume that the space X is separable, i.e., has a countable dense subset. Then to any sequence {x∗n } in X ∗ which is bounded, i.e., with kx∗n k ≤ M for some M > 0 there exists a weak-∗ convergent subsequence {x∗nk } . Weak-∗ convergent means that there exists x∗ ∈ X ∗ such that x∗nk (x) → x∗ (x) as k → ∞ for all x ∈ X. Moreover, kx∗ k ≤ M . Proof. Let x1 , x2 , . . . be a countable dense subset of X. Since {x∗n } is a bounded sequence we know that all the sequences x∗1 (x1 ), x∗2 (x1 ), x∗3 (x1 ), . . . x∗1 (x2 ), x∗2 (x2 ), x∗3 (x2 ), . . . .. . are bounded. We can therefore find convergent subsequences x∗n11 (x1 ), x∗n12 (x1 ), x∗n13 (x1 ) . . . x∗n21 (x2 ), x∗n22 (x2 ), x∗n23 (x2 ) . . . JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 104 .. . with the property that the sequence n(k+1)1 , n(k+1)2 , . . ., is a a subsequence of nk1 , nk2 , . . .. It is then clear that the tail nkk , n(k+1)(k+1) , . . . of the diagonal sequence n11 , n22 , . . . is a subsequence of nk1 , nk2 , . . . and hence that for all k ≥ 1 the sequence x∗n11 (xk ), x∗n22 (xk ), x∗n33 (xk ) . . . is convergent. Now let x ∈ X be any element of the Banach space then |x∗npp (x) − x∗nqq (x)| ≤ |x∗npp (x) − x∗npp (xk )| + |x∗nqq (x) − x∗nqq (xk )| + |x∗npp (xk ) − x∗nqq (xk )| ≤ 2M kx − xk k + |x∗npp (xk ) − x∗nqq (xk )|. Since {xk } is dense we conclude that x∗npp (x) is a Cauchy sequence for all x ∈ X. Hence x∗npp (x) is a convergent sequence for all x ∈ X. Define x∗ by x∗ (x) = limp→∞ x∗npp (x). Then x∗ is clearly a linear map and |x∗ (x)| ≤ M kxk. Hence x∗ ∈ X ∗ and kx∗ k ≤ M . B.2 COROLLARY (Banach-Alouglu on Hilbert spaces). If {xn } is a bounded sequence in a Hilbert space H (separable or not) then there exists a subsequence {xnk } that converges weakly in H to an element x ∈ H with kxk ≤ lim inf n→∞ kxn k. Proof. Consider the space X which is the closure of the space spanned by xn , n = 1, 2, . . .. This space X is a separable Hilbert space and hence is its own dual. Thus we may find a subsequence {xnk } and an x ∈ X such that xnk → x weakly in X. If y ∈ H let y 0 be its orthogonal projection onto X. We then have lim(xnk , y) = lim(xn , y 0 ) = (x, y 0 ) = (x, y). k n Thus xnk → x weakly in H. C Proof of the min-max principle In this section we give the proof of Theorem 4.12. JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 105 The operator A is bounded from below, i.e., A ≥ / − αI. In fact, from (13) we may choose α = −µ1. We first note that since vectors in D(Q) may be approximated in the k · kα norm by vectors in D(A) we may write µn = µn (A) = inf max Q(φ) : M ⊆ D(Q), dimM = n . φ∈M, kφk=1 In particular, it is no loss of generality to assume that A is already the Friederichs’ extension. It is clear that the sequence (µn ) is non-decreasing. We shall prove several intermediate results, which we formulate as lemmas. C.1 LEMMA. If for some m ≥ 1 we have µm < µm+1 then µ1 is an eigenvalue of A. Proof. Our aim is to prove that there is a unit vectors ψ ∈ D(Q) such that Q(ψ) = µ1 . It then follows from Problem 3.10 that ψ is an eigenfunction of A with eigenvalue µ1 . We may assume that µ1 = µm < µm+1 . We choose a sequence (Mn ) of m-dimensional spaces such that max φ∈Mn , kφk=1 Q(φ) ≤ µ1 + 2−4−n (µm+1 − µ1 ). We claim that we can find a sequence of unit vectors ψn ∈ Mn , n = 1, 2, . . . such that kψn − ψn+1 k ≤ 2−n . (86) In particular, the sequence is Cauchy for the norm k·k. We choose ψn inductively. First ψ1 ∈ M1 is chosen randomly. Assume we have chosen ψn ∈ Mn . If ψn ∈ Mn+1 we simply choose ψn+1 = ψn . Otherwise dim span({ψn } ∪ Mn+1 ) = m + 1 e ≥ and hence we can find a unit vector ψe ∈ span({ψn+1 } ∪ Mn+1 ) such that Q(ψ) µm+1 . In particular, we cannot have ψe ∈ Mn or ψe ∈ Mn+1 . We may write ψe = u1 − u2 , where u1 = λψ1 , λ 6= 0 and u2 ∈ Mn+1 \ {0}. We therefore have µm+1 ≤ Q(u1 − u2 ) = 2Q(u1 ) + 2Q(u2 ) − Q(u1 + u2 ) ≤ 2(µ1 + 2−4−n (µm+1 − µ1 ))(ku1 k2 + ku2 k2 ) − Q(u1 + u2 ) ≤ 2(µ1 + 2−4−n (µm+1 − µ1 ))(ku1 k2 + ku2 k2 ) − µ1 ku1 + u2 k2 = µ1 ku1 − u2 k2 + 2−3−n (µm+1 − µ1 )(ku1 k2 + ku2 k2 ) = µ1 + 2−3−n (µm+1 − µ1 )(ku1 k2 + ku2 k2 ) JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 106 where the last inequality follows since Q(φ) ≥ µ1 kφk2 for all ψ ∈ D(Q). We can rewrite this as 2n+3 ≤ ku1 k2 + ku2 k2 . (87) Since both u1 , u2 are non-zero we may use the geometric inequality u1 u ku1 − u2 k 2 − ku1 k ku2 k ≤ 2 max{ku1 k, ku2 k} . Combining this with (87) and recalling that ku1 − u2 k = 1 we obtain 2 u1 u2 8 −n ku1 k − ku2 k ≤ ku1 k2 + ku2 k2 ≤ 2 and (86) follows with ψ2 = u2 /ku2 k. Since (ψn ) is Cauchy for the norm k · k we have ψ ∈ H such that ψn → ψ for n → ∞ . In particular, ψ is a unit vector. We will now prove that ψ ∈ D(Q) and that Q(ψ) = µ1 thus establishing the claim of the lemma. Since Q(ψn ) → µ1 as n → ∞ we have that kψn k2α = (α + 1) + Q(ψn ) → α + 1 + µ1 as n → ∞. In particular, kψn kα is bounded. Since D(Q) is a Hilbert space with the inner product (φ1 , φ2 )α = (α + 1)(φ1 , φ2 ) + Q(φ1 , φ2 ) we conclude from the Banach-Alaoglu Theorem for Hilbert spaces Corollary B.2 that there is a subsequence (ψnk ) that converges weakly in D(Q) to some ψ 0 ∈ D(Q). We must have ψ = ψ 0 . In fact, for all φ ∈ H we have a continuous linear functional on D(Q) given by D(Q) 3 u 7→ (φ, u) ∈ C. Simply note that |(φ, u)| ≤ kφkkuk ≤ kφkkukα . Thus for all φ ∈ H we have (φ, ψ 0 ) = lim (φ, ψnk ) = (φ, ψ). k→∞ Hence ψ ∈ D(Q). We also have that kψk2α = lim (ψ, ψnk )α ≤ kψkα lim kψnk kα k→∞ k→∞ and thus kψkα ≤ lim kψnk kα = α + 1 + µ1 . k→∞ JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 107 Therefore Q(ψ) = kψk2α − (α + 1)kψk2 = kψk2α − (α + 1) ≤ µ1 . Since the opposite inequality Q(ψ) ≥ µ1 holds for all unit vectors in D(Q) we finally conclude that Q(ψ) = µ1 . By induction on k we will show that if µK < µK+1 for some K ≥ k then µ1 , . . . , µk are eigenvalues for A counted with multiplicities. If k = 1 this is simply Lemma C.1. Assume the result has been proved for k ≥ 1 and that µK < µK+1 for some K ≥ k + 1. By the induction assumption we know that µ1 , . . . , µk are eigenvalues for A counted with multiplicities. Let φ1 , . . . , φk be corresponding orthonormal eigenvectors. Consider the space Vk = span{φ1 , . . . , φk }⊥ . Since A is symmetric it will map Vk ∩ D(A) into Vk and the restriction Ak of A to Vk ∩ D(A) is the operator corresponding to the restriction Qk of the quadratic form Q to Vk ∩ D(Q). That µk+1 is an eigenvalue of Ak and hence an additional eigenvalue of A (counted with mulitplicity) follows from Lemma C.1 and the following claim. C.2 LEMMA. µn (Ak ) = µn+k (A). Proof. If M is any n + k-dimensional subspace of D(Q) then the projection of span{φ1 , . . . , φk } onto M is at most k-dimensional and hence M ∩ Vk must have dimension at least n. Thus max φ∈M, kφk=1 Q(φ) ≥ max φ∈M ∩Vk , kφk=1 Q(φ) ≥ µn (Ak ). Thus µk+n (A) ≥ µn (Ak ). To prove the opposite inequality note that if φ = φ1 +φ2 ∈ D(Q) with φ1 ∈ Vk and φ2 ∈ span{φ1 , . . . , φk } we have Q(φ) = Q(φ1 ) + Q(φ2 ). (88) JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 108 We first show that µ1 (Ak ) ≥ µk (A). Assume otherwise that µ1 (Ak ) < µk (A) we can then find a unit vector φ0 ∈ Vk such that Q(φ0 ) < µk (A). Let j be the largest integer such that µj (A) ≤ µ1 (Ak ) (this is certainly true for j = 1). Then j < k. If we consider the j + 1-dimensional space M = span{φ1 , . . . , φj , φ0 } we see from (88) that µj+1 (A) ≤ max φ∈M, kφk=1 Q(φ) ≤ µ1 (Ak ) which contradicts the fact that j was the largest integer with this property. Hence we must have µ1 (Ak ) ≥ µk (A). From (88) we find that if M 0 is any n-dimensional subspace of Vk for n ≥ 0 we have for the n + k-dimensional subspace M = M 0 ⊕ span{φ1 , . . . , φk } that max Q(φ) = max µk (A), max Q(φ) . 0 φ∈M, kφk=1 φ∈M , kφk=1 Hence µk+n (A) ≤ max {µk (A), µn (Ak )} = µn (Ak ). The statement in the second paragraph of Theorem 4.12 follows immediately from Lemma C.2. The last statement is an easy exercise left for the reader. D Analysis of the function G(λ, Y ) in (37) If we use the inequality 2ab ≤ a2 + b2 on the last term in G(λ, Y ) we see that G(λ, Y ) ≤ (1 − λ2 ) (1 − Y 2 ) N (M − N ) (1 − λ2 ) Y 2 (N − 2) (M − N + 2) + M −2 M −2 2 2 λ (1 − λ ) (M − N )N +2 λ2 Y 2 + 2 Y 2 1 − Y 2 + . M −2 We see that this expression is a quadratic polynomial p(x, y) in the variables x = λ2 , y = Y 2 . Straightforward calculations show that ∂2 N (M − N ) ∂ 2 ∂2 M −N p(x, y) = −2 , p(x, y) = −4, p(x, y) = 4 2 2 ∂x M −2 ∂y ∂x∂y M −2 and that ∂ p(x, y)(2/M, N/M ) = 0, ∂x ∂ p(x, y)(2/M, N/M ) = 0. ∂y JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 109 In particular, ∂2p ∂2p − ∂x2 ∂y 2 ∂2p ∂x∂y 2 =8 (M − N ) M (N − 2) ≥ 0. (M − 2)2 This shows (by the second derivative test) that p(x, y) is maximal for (x, y) = (2/M, N/M ). Thus G(λ, Y ) ≤ p(2/M, N/M ) = E N (M − N + 2) = G((2/M )1/2 , (N/M )1/2 ). M Results on conjugate linear maps Recall that a conjugate linear map C : H → K between complex vector spaces H and K is a map such that C(αu + βv) = αC(u) + βC(v) for all u, v ∈ H and α, β ∈ C. We will concentrate on the situation where H and K are Hilbert spaces. The map J : H → H∗ given by J(φ)(ψ) = (φ, ψ) (see also Remark 1.2) is conjugate linear. The adjoint of a conjugate linear map C : H → K is the conjugate linear map ∗ C : K → H defined by (Ch, k)K = (C ∗ k, h)H , for all h ∈ H, k ∈ K (89) The map J : H → H∗ is anti-unitary meaning J ∗ J = IH and JJ ∗ = IH∗ . (90) E.1 PROBLEM. Show that (89), indeed, defines a conjugate linear map C ∗ and show the identities in (90). E.2 THEOREM (Conjugate Hermitian and anti-Hermitian maps). Let C : H → H be a conjugate linear map such that C ∗ C is trace class. If C : H → H is a conjugate Hermitian map, i.e., a conjugate linear map satisfying C ∗ = C then H has an orthonormal basis of eigenvectors for C and all the eigenvalues are non-negative (in particular real). JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 110 If C : H → H is conjugate anti-Hermitian, i.e., a conjugate linear map satisfying C ∗ = −C then ker(C) is a closed subspace of H and the space ker(C)⊥ has an orthonormal basis u1 , u2 . . . such that Cu2i = λi u2i−1 , Cu2i−1 = −λi u2i , where λi > 0, i = 1, 2, . . .. Proof. The operator C ∗ C is a linear Hermitian positive semi-definite map since (C ∗ Cu, u) = (Cu, Cu) ≥ 0. Since C ∗ C is trace class it can be diagonalized in an orthonormal basis and each non-zero eigenvalue has finite multiplicity. Moreover, if C is Hermitian or anti-Hermitian C maps the eigenspace of C ∗ C into itself. Assume that u is a normalized eigenvector for C ∗ C, i.e., Correction since June 24: Invariance C ∗ Cu = λ2 u for some λ ≥ 0. C added Correction We consider first the case C ∗ = C. We then have (C + λ)(C − λ)u = (C 2 − λ2 )u = 0. Hence either (C − λ)u = 0 or w = (C − λ)u 6= 0 and (C + λ)w = 0. of eigenspaces under June in 24: since u, w eigenspace of C ∗ C added We have thus found one eigenvector (either u or w) for C, which belong to the eigenspace of C ∗ C with eigenvalue λ. We can always assume the eigenvalue is nonnegative by eventually multiplying the eigenvector by i and using the conjugate linearity, i.e., Ciw = −iCw = λiw if Cw = −λw. We will now show that C maps the orthogonal complement of this eigenvector into itself. We may therefore finish the proof by a simple induction over the dimension of the eigenspace of C ∗ C with eigenvalue λ. Correction Assume that u is an eigenvector for C and that v ⊥ u. We must show that Cv ⊥ u. This is easy since June of H by replaced dimension eigenspace (Cv, u) = (Cu, v) = 0, where the first identity follows since C ∗ = C and the second identity follows since U is an eigenvector of C and v ⊥ u. Consider next the case when C ∗ = −C. If the eigenvalue λ2 of C ∗ C vanishes then Cu = 0, since (Cu, Cu) = (C ∗ Cu, u) = 0. We can therefore choose u as one 24: Dimension of JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 111 of the basis vectors. We may then proceed as before and show that C maps the orthogonal complement of u into itself and then reduce the problem to a space of lower dimension. If λ > 0 then kCuk = λ and we define the unit vector w = λ−1 Cu. Then Cu = λw and Cw = λ−1 C 2 u = −λ−1 C ∗ Cu = −λ1 λ2 u = −λu. Moreover, (w, u) = λ−1 (Cu, u) = λ−1 (C ∗ u, u) = −λ(Cu, u) = −(w, u). Thus (w, u) = 0. Thus u and w can be the first two vectors in the orthonormal basis. If we can now show that C maps the orthogonal complement of {u, w} into itself we can again finish the proof by induction. Assume therefore that v ⊥ u and v ⊥ w. We have (Cv, u) = (C ∗ u, v) = −(Cu, v) = −(λw, v) = 0 and (Cv, w) = (C ∗ w, v) = −(Cw, v) = (λu, v) = 0. and hence Cv ⊥ {u, w}. Proof of Lemma 8.14. To each element f ∈ h ∧ h we may associate a conjugate linear map Cf : h → h by (φ, Cf ψ)h = (φ ∧ ψ, f )h∧h for all φ, ψ ∈ h. Then (φ, Cf∗ ψ)h = (ψ, Cf φ)h = (ψ ∧ φ, f )h∧h = −(φ, Cf ψ)h . Hence Cf is a conjugate anti-Hermitian map. We may then choose an orthonormal basis u1 , . . . , u2r , u2r+1 , . . . , uM for h such that u2r+1 , . . . , uM are in the kernel of Cf and u1 , . . . , u2r are as described in Theorem E.2. We claim that f= r X i=1 λi u2i−1 ∧ u2i . JPS— Many Body Quantum Mechanics Version corrected July 6, 2007 112 This follows from (φ ∧ ψ, f ) = (φ, Cf ψ) = n X n X (φ, ui )(ui , Cf (uj , ψ)uj ) i=1 j=1 = = = n X n X (φ, ui )(ui , Cf uj )(ψ, uj ) i=1 j=1 r X (φ, u2i−1 )(u2i−1 , Cf u2i )(ψ, u2i ) + (φ, u2i )(u2i , Cf u2i−1 )(ψ, u2i−1 ) i=1 r X λi ((φ, u2i−1 )(ψ, u2i ) − (φ, u2i )(ψ, u2i−1 )) i=1 = (φ ∧ ψ, r X λi u2i−1 ∧ u2i ). i=1 F The necessity of the Shale-Stinespring condition TO COME